A delegation from the European Parliament has begun a three-day mission to Bosnia and Herzegovina to examine foreign interference in the country’s political institutions, electoral processes and public information environment.
The visit, taking place from 26th to 28th October, comes three weeks after parliamentary elections exposed persistent divisions over Bosnia’s political direction and its prospects of joining the European Union.
The mission is being conducted by Parliament’s Special Committee on the European Democracy Shield, chaired by French MEP Nathalie Loiseau. Its purpose is to assess vulnerabilities to disinformation, cyberattacks, economic pressure and other forms of hybrid interference, and to identify measures that could strengthen Bosnia’s institutional resilience.
According to the European Parliament’s announcement, the delegation will meet government officials, parliamentary representatives, civil society organisations, journalists and researchers. Joining Loiseau are Finnish MEP Mika Aaltola, Portugal’s Ana Catarina Mendes and Romania’s Cristian Terheş.
The investigation comes at a politically sensitive moment. Bosnia’s general election on 4th October confirmed the persistence of competing political priorities between the country’s constituent entities, with pro-Russian politician Milorad Dodik’s Alliance of Independent Social Democrats retaining considerable support in Republika Srpska.
As EU Today reported following the election, parties favouring closer European integration also performed strongly, particularly in the Bosniak-Croat Federation. The resulting political balance will influence the formation of new administrations and their ability to implement reforms demanded by Brussels.
From electoral interference to institutional security
The European Parliament’s investigation extends beyond conventional disinformation campaigns.
MEPs intend to examine how foreign actors may exploit weaknesses in electoral administration, cybersecurity, political financing and the country’s wider economic relationships. They will also consider the role of digital platforms in distributing manipulated information and the legislative gaps that could make interference difficult to identify or prosecute.
The distinction matters in Bosnia, where political influence cannot be separated from the country’s complicated constitutional arrangements.
Under the Dayton peace settlement, the state operates through institutions that distribute authority between the Federation of Bosnia and Herzegovina, Republika Srpska and the central government. Political disagreements can therefore obstruct decisions requiring cooperation between different levels of administration.
External actors seeking influence may exploit those divisions, although identifying a political relationship with a foreign government is not, by itself, evidence of unlawful interference.
The European Union has repeatedly expressed concern about Russia’s activities in the Western Balkans, particularly its relationships with political forces opposing closer integration with European institutions. Chinese investment and economic relationships also raise questions about strategic dependence, although legitimate commercial activity must be distinguished from coercion or covert political operations.
The delegation’s task is to establish which vulnerabilities can be substantiated and whether existing safeguards are adequate.
Accession negotiations remain incomplete
Bosnia received EU candidate status in December 2022. In March 2024, European leaders agreed in principle to open accession negotiations, but progress has remained conditional on institutional and legal reforms.
Following the October election, EU foreign policy chief Kaja Kallas and Enlargement Commissioner Marta Kos issued a joint statement calling for the prompt formation of new administrations and progress on judicial reform.
The Commission also highlighted shortcomings identified by international election observers, including technical difficulties, complicated voting procedures and concerns about ballot secrecy.
Those findings do not automatically establish foreign interference. They do, however, demonstrate why the integrity of electoral institutions remains central to Bosnia’s relationship with Brussels.
The initiative is intended to improve cooperation on election security, information manipulation, crisis management and institutional protection before candidate countries enter the Union.
EU Today has previously examined the European Democracy Shield’s proposed mechanisms, including closer coordination with technology companies and improved systems for identifying foreign information operations.
Bosnia now offers a practical opportunity to examine whether those instruments can operate effectively in a country where political authority is fragmented and accession remains incomplete.
The delegation’s findings could inform recommendations on institutional cooperation, election safeguards and support for independent media. Any substantial improvement, however, will depend on the willingness of Bosnia’s authorities to implement reforms and enforce existing legislation.
For Brussels, the immediate issue is whether its emerging framework for democratic resilience can provide practical protection to candidate countries while respecting their political sovereignty and the freedoms it is intended to defend.
The violence accompanying France’s October protests has revealed how foreign influence networks can exploit domestic grievances to undermine governments and European support for Ukraine. New evidence suggests Russian-linked operations were active before the riots, raising questions about whether information manipulation is moving beyond propaganda towards the mobilisation of political unrest.
France’s violent protests have exposed a vulnerability in European security that conventional policing cannot easily address: the ability of foreign information networks to exploit domestic political crises, amplify social divisions and transform legitimate grievances into instruments of geopolitical pressure.
President Emmanuel Macron has acknowledged foreign manipulation of information surrounding the disturbances of 17–18 October but insists there is no evidence that foreign actors organised the violence itself. His assessment reflects the evidence currently available, although new investigations indicate that Russian-linked networks were actively exploiting the protest movement weeks before the worst disorder began.
On 24 October, France deployed approximately 70,000 police officers and gendarmes to prevent further violence. Yet the security operation addresses only the physical consequences of a crisis whose online dimension extends across national borders.
Researchers examined 547,471 posts and articles across nine languages. They identified 639 coordinated messages attributed with high confidence to Russia’s Doppelgänger operation, generating approximately 7.4 million views in the six weeks preceding 17 October.
The campaign used genuine concerns about living costs and public expenditure to promote a particular political argument: money allocated to Ukraine was being taken from European citizens.
This was not a random selection of grievances. France’s budget difficulties, public-service pressures and disagreements over defence expenditure provided material through which Moscow could seek to weaken support for Kyiv.
The researchers found that the operation targeted pro-European and pro-Ukraine politicians across the political spectrum. It praised selected proposals associated with Jean-Luc Mélenchon while also promoting nationalist narratives, illustrating a strategy driven less by ideological consistency than by opposition to existing European policies.
From manipulating opinion to encouraging mobilisation
Russian-linked channels were not merely commenting on existing demonstrations. They were circulating protest dates, identifying organisers and encouraging French audiences to follow the example of demonstrators in Germany.
On 28 September, the International Reporters network published an appeal urging French audiences to take to the streets. Other channels subsequently reproduced information about planned school blockades and demonstrations.
In one instance, a channel identified as Gavroche published rhetoric encouraging an attack on France’s Education Ministry.
Such activity goes beyond retrospective propaganda. It attempts to influence participation and behaviour while events are developing. Nevertheless, evidence that a network encouraged protests does not establish that it controlled the organisers or directed subsequent violence.
The original demonstrations arose from domestic grievances over schools, prices and government policy. The significance of the foreign activity lies in attempts to redirect and intensify an existing movement.
How France’s crisis was rewritten abroad
Once violence erupted, the information campaign acquired an international audience.
Alliance4Europe found that claims blaming migrants or Islam for the disturbances appeared in approximately one in 33 French-language posts, compared with nearly one in five English-language posts.
False allegations that immigrants had burned Limoges town hall circulated substantially more frequently in English than in French.
The same report identified 1,018 articles distributed through 28 websites associated with the Russian Pravda network. Their rapid republication across languages enabled a common narrative to reach audiences far beyond France.
The operation also presents a longer-term technological problem. Repeated publication of manipulated accounts can contaminate material subsequently retrieved or reproduced by artificial-intelligence systems, allowing false narratives to persist beyond the original campaign.
The political effects were visible when President Donald Trump and other foreign leaders portrayed France’s unrest principally through immigration and cultural conflict.
Speaking in Qatar on 21 October, Macron rejected these portrayals and criticised the circulation of fabricated AI-generated images. His remarks identified foreign manipulation after the riots but stopped short of alleging foreign responsibility for their organisation.
The Heritage Foundation and transatlantic networks
The transcript raises a more substantial allegation: that the American Heritage Foundation and Russia’s FSB have participated in a coordinated effort to destabilise France.
There are documented connections between the Heritage Foundation and political institutions associated with Hungary’s former prime minister Viktor Orbán.
These relationships demonstrate the development of transatlantic political networks capable of circulating common ideas about sovereignty, migration and European integration.
They do not establish operational cooperation with Russian intelligence. Nor has credible evidence emerged connecting Heritage to the organisation or financing of France’s October riots. Ideological convergence, institutional cooperation and participation in a clandestine operation require separate evidential assessments.
The implications for European security
The European Union’s response has exposed a regulatory problem.
Under the EU’s Digital Services Act, major online platforms must address systemic risks arising from their services, including the amplification of illegal content and threats to public security.
These measures targeted unlawful online activity, particularly incitement to violence and looting. They did not constitute findings that the suspended accounts belonged to foreign intelligence networks.
The European Commission has likewise refrained from activating emergency measures under the Digital Services Act without sufficient evidence connecting foreign manipulation to the violence.
The broader security concern is that foreign operations can exploit the interval between the emergence of social unrest and the establishment of reliable information about its causes.
Russia need not organise an entire protest movement to advance its strategic interests. By amplifying domestic grievances, promoting hostility towards support for Ukraine and circulating distorted accounts internationally, it can attempt to convert France’s internal political difficulties into pressure on European policy.
With France approaching its 2027 presidential election, these operations warrant continuing scrutiny. The central investigative question is no longer simply whether foreign networks spread false information after violence occurred, but whether their activities before and during demonstrations materially influenced mobilisation, radicalisation or political responses.
The evidence currently demonstrates organised manipulation and attempts at amplification. Establishing direct responsibility for violence would require a further evidential step, involving financial records, communications, operational instructions or connections with identifiable perpetrators.
France’s experience illustrates how contemporary information warfare can operate within a democratic society without necessarily directing events on the ground. Its immediate objective may be to alter the political interpretation of a crisis; its longer-term consequence may be to influence the policies adopted in response.
Italian authorities are examining whether foreign interference helped intensify violent protests across the country, as Defence Minister Guido Crosetto warns that social media is being used as an instrument of hybrid warfare. Prime Minister Giorgia Meloni has introduced emergency security measures and deployed additional troops to Milan, while further demonstrations are planned for Saturday, 24 October.
Italy is investigating the possibility that foreign actors, including Russian operatives, helped amplify the violent unrest that swept Milan and other Italian cities on 17–18 October. The inquiry introduces a national security dimension to demonstrations initially driven by rising living costs, dissatisfaction with public services and broader economic grievances.
According to information disclosed about the government’s intelligence inquiries, officials working under Prime Minister Giorgia Meloni have requested assessments of whether foreign provocateurs contributed to the disorder. No public evidence has established that Moscow organised the violence, and Meloni’s office has declined to comment on the reported intelligence discussions.
The developments follow EU Today’s earlier coverage of the Milan disturbances, which documented extensive looting, attacks on police and the detention of hundreds of young people, many of them minors. Since then, Rome has moved beyond emergency consultations, approving legislation that expands police powers and introduces tougher criminal penalties for violent demonstrations.
On Saturday, 24 October, Italian security forces are preparing for another wave of protests, with additional military personnel, drones and specialist police units deployed in several major cities. The immediate question is whether the government can prevent further disorder while establishing how much of the unrest was locally organised and whether foreign information operations influenced its development.
Crosetto warns of hybrid warfare through social media
Defence Minister Guido Crosetto introduced the hybrid warfare argument during a Defence Procurement Forum in Rome on Tuesday, 20 October. In remarks recorded by Italian public broadcaster Rai, he described the disturbances as the culmination of a form of conflict being conducted through TikTok, YouTube and other social media platforms.
Crosetto argued that young Italians were undergoing what he called a form of counter-education outside their families and schools, absorbing political attitudes and values increasingly hostile to democratic institutions. He criticised Western governments for failing to recognise the consequences of this process, although he did not publicly present intelligence establishing Russian responsibility for the riots.
His comments also exposed differences within the governing coalition. Deputy Prime Minister Matteo Salvini questioned suggestions of Russian involvement, arguing that the immediate priority was identifying violent groups and preventing their movement into Milan.
Russian information operations under scrutiny
European investigators have identified activity by established Russia-linked disinformation networks, including Operation Overload and Doppelgänger, which have circulated and amplified material concerning the unrest.
These operations have previously used manipulated content, misleading narratives and coordinated online distribution to influence European audiences. In the present case, investigations into the online amplification of the violence have identified foreign exploitation of the disturbances, rather than proof that foreign governments organised them.
The European Commission has consequently declined to initiate emergency intervention against social media platforms, citing insufficient evidence linking foreign manipulation directly to the organisation of the protests.
France has taken a similarly cautious position. President Emmanuel Macron acknowledged that foreign actors had manipulated information after the violence but said there was no established evidence that they caused the disturbances. French Foreign Minister Jean-Noël Barrot has separately identified hostile narratives originating from Russia, China and Iran.
For European security authorities, the principal investigative question concerns the relationship between genuine domestic grievances and externally coordinated information campaigns. Foreign networks can exploit political divisions without necessarily directing the organisations or individuals involved in street violence.
Meloni introduces tougher security legislation
Meanwhile, Meloni’s government has proceeded with a substantial legislative response. Following cabinet approval on 20 October, Decree-Law No. 190 of 22 October 2026 entered into force on 23 October.
The legislation introduces penalties of two to five years’ imprisonment and fines of €10,000 to €20,000 for specified forms of organised violence during public demonstrations. Those organising or encouraging qualifying violent groups, including through digital communications, face potential prison terms of four to twelve years. It also expands provisions concerning juvenile offenders and preventive policing. The decree must undergo parliamentary consideration within 60 days.
Separately, the Italian Defence Ministry confirmed the deployment of 500 additional soldiers to Milan under Operation Strade Sicure until 31 December. The soldiers will reinforce security around sensitive locations and support existing police operations without acquiring additional judicial policing powers.
New demonstrations test the government’s response
Security arrangements have been intensified ahead of demonstrations scheduled for Saturday, 24 October, in Milan, Rome, Turin, Bologna, Florence and other cities.
According to the latest security preparations, Rome is deploying approximately 1,500 officers, while Milan is preparing a combined police and military presence supported by aerial surveillance.
On Friday, Italian investigators conducted searches targeting approximately 100 people suspected of involvement in the previous weekend’s disturbances, including 51 minors. Authorities are also monitoring online calls for renewed protests, some organised by groups whose leadership and membership remain unclear.
The unrest presents Meloni with two parallel challenges: containing violence and determining whether hostile foreign networks have played a significant role in its escalation. Italy’s investigations may help establish whether the events represent primarily a domestic public-order crisis, the exploitation of existing tensions by foreign actors, or a combination of both.
For Brussels, the outcome could influence future enforcement of the EU’s Digital Services Act and its wider response to foreign information manipulation. For now, the evidence supports concern about Russian amplification of European unrest, but not a definitive conclusion that Moscow orchestrated the riots.
President Trump, who has continued to falsely claim that he won the 2020 election, has been relentless in his efforts to sow doubts about the integrity of American voting.
Sergei Kiriyenko, first deputy chief of staff of the Russian Presidential Administration, set the goals for the Ukraine strand of the operation, launched in early 2023 under the codename "Uganda." Alexander Kharichev, now head of the Kremlin's Directorate for Monitoring and Analysis of Social Processes, oversaw it, according to hacked correspondence published by The Insider on 21 October.
Doppelgänger disguises Russian narratives as local news through cloned media sites, fake accounts, paid ads, and fabricated videos. The FBI named Kiriyenko among its orchestrators in 2024, and its contractor, the Social Design Agency (SDA), run by Ilya Gambashidze, is still expanding, now targeting AI chatbots. The new files go inside the Ukraine campaign and show the operation reporting even exposed fakes to the Kremlin as achievements.
Telegram channels posing as Ukrainian
An internal briefing on narratives says the project ran its Telegram channels on the principles of agent work and psychological operations, the Russian version of the investigation shows. Early on, the team decided to disguise all its outlets as Ukrainian and to run its campaigns without direct Russian propaganda. Every channel carried Ukrainian news to look unconnected to Russia. Dropping pro-Ukrainian themes, the briefing warned, would eventually expose the client and wreck the whole plan.
Explore further
Russia-linked bot networks kept their weekly rhythm after drone strikes on three Yandex data centers—disinformation monitor LetsData
Four goals and a can of stew
Kiriyenko set four goals: discredit Ukraine's military and political leadership, split its elites, demoralize its army, and disorganize its population. The SDA prepared the project's pitch and carried out the work.
Political strategist Alexander Asafov proposed fakes of his own: in May 2023, that soldiers were bribing commanders to avoid the front, and in January 2024, that Ukraine was pulling air defenses back to Kyiv to protect the authorities.
After an Il-76 military transport crashed in Belgorod Oblast on 24 January 2024, Moscow claimed it had been carrying Ukrainian prisoners of war. Kharichev's subordinate Yevgeny Kazakov, chairing a meeting in his place, proposed sending a fake to the widows of Ukrainian soldiers: "They died for an expensive Il-76—the Russians have few of those." The idea was dropped.
The same meeting produced another fake, and three days later the team launched it: that recruitment centers had sent widows a can of stewed meat with a little flag as compensation. The same producers also oversaw letters falsely attributed to Deputy Defense Minister Nataliia Kalmykova, the Russian version says.
Debunks filed as results
The SDA collected coverage of its work in a folder for the Kremlin. It filed the "Zelensky's yacht" story, which JD Vance, then a US senator and now US Vice President, echoed, together with its debunk. The report also included a TikTok campaign about former Defense Minister Oleksii Reznikov's family that the BBC had already exposed.
The Insider concludes that the team has had more failures than successes. The US sanctioned Gambashidze and the SDA in March 2024, and the UK followed in October.
In 2016, the HBO comedian and political commentator John Oliver produced a terrific segment on how TV pundits routinely cite garbage science. After showing clips of newscasters and commentators making absurd claims such as, "A new study showing that drinking a glass of red wine is just as good as spending an hour at the gym," and The TodayShow's Al Roker commenting, "You find the study that sounds best to you, and you go with it," Oliver commented, "No! No, no, no, no, no, no, no! You don't just get to cherry-pick the parts that justify what you were going to do anyway."
Then Oliver added a useful rule: "If they are going to keep saying 'a study says,' they should have to provide sourcing and context or not mention it at all." (Though the segment specifically targeted the misuse of "scientific" studies, all the same principles apply to the social sciences.)
What's so noteworthy about that 2016 segment is that in his other monologues, on issues ranging from SWAT teams to homelessness, Oliver has routinely ignored the principles he laid out. In his overly simplistic takes on complex policy issues (often designed to dunk on his ideological opponents for comedic effect), he often fails to provide appropriate context and appears to be guilty of choosing research "that sounds best," as Roker put it, or at least that supports whatever point he's determined to make.
Oliver is one of America's most influential pundits, so he should be held to the highest of journalistic standards. And Last Week Tonight is indeed more diligent than many TV news programs. The show employs researchers who interview the authors of the studies it cites. Oliver is a comic, not an academic, so we can excuse him for not posting footnotes, and he routinely cites his sources on screen. When we emailed the show's producers to point out problems with their research, to their credit, they sent back a detailed response that we'll quote from in this article. (In a few cases, they pointed to Reason articles using some of the same studies, albeit with full citations and more context. In cases in which their criticisms were correct, we've annotated the original articles.)
Many of the show's researchers are former reporters and editors who abide by journalistic standards. Nevertheless, the show has a long track record of citing social science findings without the appropriate context, massaging them to support whatever point Oliver is trying to make.
Rent
In his excellent 2016 segment on scientific studies, Oliver noted: "There is a lot of bullshit currently masquerading as science. Not all scientific studies are equal. Some may appear in less-than-legitimate scientific journals, and others may be subtly biased because of scientists feeling pressured to come up with eye-catching positive results."
Cut to a 2022 segment titled "Rent," and we find Oliver citing a study that his wiser self would have characterized as "bullshit."
"Right now, there is not a single county in the U.S. where a worker earning minimum wage can afford a modest two-bedroom rental home," he said. He was citing a study designed to produce an "eye-catching" result. It was published not by a reputable journal, but by the National Low Income Housing Coalition, an advocacy group that lobbies for government spending on housing and rent subsidies, including expanding the National Housing Trust Fund and increasing federal appropriations to the Department of Housing and Urban Development.
The biases are not particularly subtle. Why would a minimum-wage worker need a two-bedroom home? In the study, "can afford" doesn't mean "have enough money to buy the cheapest available," as most English speakers would interpret it. The authors use it to mean renting a two-bedroom home at the 40th percentile price point of the local market (that is, a little cheaper than the median unit), using no more than 30 percent of their income, excluding government benefits, pensions, or other nonearned support.
Oliver also didn't mention who these minimum-wage workers are. In 2025, according to the Current Population Survey, a third of hourly workers paid the federal minimum or less lived in a parent's or grandparent's home, a fifth were married and living with a spouse, and only one in eight lived alone.
When we asked the show's producers why they failed to disclose that their source was an advocacy group seeking government funds, they pointed to the onscreen credit and responded: "We trust our viewers' intelligence enough to think they might be able to read it and conclude that the National Low Income Housing Coalition is interested in building low-income housing, nationwide, possibly even as part of a coalition."
I don't share their confidence that most viewers can read the onscreen source line and tell from the organization's name whether it's an advocacy group. There's a double standard here: When Oliver doesn't like what a study says, he's not content to just trust his viewers; he makes it explicit, as he did with the "ethanol lobbying groups" mentioned above.
Oliver's producers wrote us: "When we rely on studies from advocacy groups, we look at the underlying data and speak to researchers to better understand it; we discard many studies as a result."
This one belonged in the trash.
Sports Betting
In a 2025 segment, Oliver cited an eye-catching statistic about how sports betting was driving a spike in bankruptcies. "Unsurprisingly, the rise of online sports betting has been found to correlate with a rise in gambling harms, with one study showing that in states that allowed it, there was a 28 percent increase in bankruptcies, translating to about 100,000 extra bankruptcies a year."
I'll give Oliver credit for saying "correlate with" rather than "cause," but the rest of that sentence ("in states that allowed it, there was a 28 percent increase") implies direct causality.
The bigger problem is that the "100,000" figure is clearly wrong. When I contacted the study's authors, they told me they don't know where that figure came from, and it doesn't appear anywhere in the study. Last Week Tonight's producers told us they pulled it from a Business Insider article, but in the video, Oliver said it came from the study.
If Oliver's researchers had done the same calculation as the authors, multiplying the per-person effect by the at-risk population, they'd have found about 30,000 extra bankruptcies a year, not 100,000, and a 6 percent to 7.5 percent increase, not 28 percent.
Oliver also left out important context: He was citing an unpublished working paper. The peer-reviewed version appeared after the Last Week Tonight segment aired, and it included substantive changes. Working papers from credible researchers, like this one, are valuable, but Oliver should have made it clear he was citing a preliminary draft circulated to other researchers for comment, not a published study.
Homelessness
In 2021, Oliver did a segment making a case for a policy called "Housing First," which includes placing more homeless people in permanent supportive housing. The problem is that building and maintaining these units is expensive, so permanent supportive housing doesn't scale easily. Oliver set out to dismiss that objection by citing research showing the alternative is even more costly.
"One study in Florida," Oliver said, "found that, just between money spent on incarceration and emergency medical treatment, their communities and local governments had spent an average of just over $31,000 per person per year, when the estimated cost of providing them with permanent supportive housing would have only been around $10,000 per year."
The Florida study's finding was an outlier. The Central Florida Commission on Homelessness commissioned it, and Housing First advocates promoted it because it produced the headline they wanted. I give Oliver credit for noting that the study "tracked a decade's worth of spending on just 107 chronically homeless people," but he left out additional context.
In 2018, a committee convened by the National Academies of Sciences, Engineering, and Medicine evaluated permanent supportive housing. It concluded that there is "no substantial evidence that [it] contributes to improved health outcomes" and "insufficient evidence" that it's cost-effective.
Studies exist on both sides of this issue, and we can debate whether permanent supportive housing is more cost-effective than the alternative, but Oliver cited one study and implied it was conclusive.
Oliver also neglected to mention that the paper was written by a consulting firm led by a supportive-housing advocate, though if you looked closely, you could see the organization's name on the screen as he delivered his monologue.
Ethanol
Oliver and I agree that the federal government shouldn't be subsidizing ethanol. But in a 2024 segment titled "Corn," Oliver pointed to a study claiming that ethanol is more harmful to the environment than gasoline and implied that only self-interested lobbyists say otherwise.
"While ethanol lobbying groups have long argued that it slashes greenhouse emissions," he said, "one recent study found that corn ethanol produced under the renewable fuel standard has a carbon footprint at least 24 percent higher than regular gasoline."
What Oliver doesn't mention is that the 2022 study he cited was highly contested. It was challenged not only by "ethanol lobbying groups" but also by the U.S. Department of Agriculture under the Biden administration. The study's claim that ethanol has a higher carbon footprint than gasoline also contradicts a 2010 Environmental Protection Agency projection that ethanol would be about 20 percent cleaner than gasoline. Argonne National Laboratory, running the government's standard model, credits ethanol with cutting greenhouse emissions by more than 500 million tons in 15 years.
The problem isn't that Oliver is wrong; it's how he presented this study. Whether ethanol mandates are good or bad for the environment is a complicated, contested issue, with strong studies on both sides (and plenty of weak ones). Plucking one from the pile and reporting its findings is misleading.
Ethanol subsidies are bad because they involve politicians using mandates to buy farmers' votes, and top-down mandates generally lead to less optimal decision-making. But it's not clear that ethanol increases carbon emissions.
SWAT Raids
In a 2014 segment on police militarization after the fatal shooting of Michael Brown in Ferguson, Missouri, Oliver said, "There has been a massive rise in both the number and the use of police department SWAT teams," and showed a news clip that stated: "The number of SWAT raids has gone up by 1,400 percent since the 1980s. An estimated 50,000 now take place every year." Instead of mocking the clip, as he often does, Oliver cited it approvingly.
So where did the claim of a 1,400 percent increase come from? It's based on a voluntary mail-in survey, which isn't a reliable way to gather data. No official count of SWAT raids exists, and almost nobody tracks them, then or now, so the figure rests on departments' self-reporting of how often they "deployed," with no shared definition of what that means.
The 1980s figures that form the baseline for the 1,400 percent increase in SWAT raids come from a mid-1990s survey, making them especially unreliable. No one was counting SWAT raids at the time, and there's no way to verify them.
Oliver followed that clip by citing another study: "Just keep that clip in mind," he said, "when I tell you that 79 percent of SWAT deployments are now for executing search warrants, and most of those were for drug investigations."
Again, there's missing context. That 79 percent figure comes from a small, nonrandom sample generated by an American Civil Liberties Union records request; the organization could get data from just 20 of America's roughly 18,000 police agencies.
When we pointed out to Oliver's producers that the data on SWAT teams were stripped of their context, they replied that since this segment aired, "We've become more diligent about critically interrogating the sourcing on studies in our shows." They also argued that more "recent studies," if anything, suggest that the studies they cited "understated the true scope of things." And they pointed to instances of Reason citing the same data.
I agree with Oliver's position on SWAT raids, but he really should be agitating for better data. We're permitting state violence without anyone keeping the detailed records that could help us understand the costs and benefits. But that doesn't excuse citing two studies based on weak data and presenting them as fact.
Tipping
In a 2025 segment on tipping, Oliver said, "Recent analyses have found that poverty rates for tipped workers are over twice as high as for nontipped workers."
Twice as high as for which nontipped workers? The analysis he's citing included all nontipped workers in the U.S., including CEOs and neurosurgeons. Oliver, who reportedly makes $30 million a year, also pulled up the nontipped average. The right way to evaluate tipping is to match tipped and untipped workers by job: Waiters have about the same poverty rate as fast-food workers, and hotel maids about the same as other cleaners. If you control for job, age, and hours, the gap closes.
It's another study designed to generate a talking point, which isn't surprising since it was published by the Economic Policy Institute, a progressive, left-of-center think tank. The organization's name appeared onscreen as Oliver delivered his monologue, but, once again, viewers would need to hit pause to look it up and figure out what the Economic Policy Institute stands for.
Oliver's 2016 segment on "scientific studies" shows that he should know better. It's possible to construct arguments by citing good research responsibly, while providing the proper context and sourcing. It's possible to follow research where it leads, instead of retrofitting it to preconceived ideas. You just have to be willing to present data that capture reality in all its messiness, even if that means it won't generate an easy laugh line.
To keep up with our video series Wrong Number featuring Aaron Brown, and to receive bonus content and more, click here to enter your email address.
The midterms are the first elections in which artificial intelligence is integrated into every major social media platform. Efforts to blunt its harms are uneven.
One in five voters plans to use A.I. to look into a race on their ballot, according a survey released this month by the American Association of Political Consultants.
From the 1918 influenza pandemic to reports of a possible laboratory accident in Irkutsk, history shows how incomplete evidence, sensationalist reporting, and geopolitical rivalry can turn medical uncertainty into a tool of political narrative-building. The Spanish flu was not named for its place of origin. It acquired its misleading name in the information environment of […]
In recent days, a Ukrainian drone strike hit Russia’s largest data center, owned by bank and technology company Sberbank. The strike came after Ukraine hit three data centers belonging to the tech giant Yandex, which is often referred to as “Russia’s Google.” After those attacks, Yandex’s traffic dropped 94%. Its users experienced heavy disruption to […]
Ask a chatbot why something happened in the war, and the answer may be written by Russia. Russia is poisoning AI chatbots with millions of fake articles so they repeat Kremlin narratives as fact, a cognitive weapon with strategic impact, Ukraine’s Unmanned Systems Forces commander Robert “Madiar” Brovdi warns on Telegram.
Russia has waged a cognitive war against Ukraine for years, and it intensified sharply with the full-scale invasion. This war goes beyond fake news and disinformation: it aims to shape the deeply ingrained beliefs that drive people’s attitudes, decisions and behavior, and its newest weapon is artificial intelligence.
Brovdi calls it a cognitive operation with strategic impact, one his forces studied among the technological risks of 2027. People increasingly ask a chatbot “why did X happen?” and take the instant answer, he explains, without asking where it comes from.
“Artificial intelligence does not know the truth. It retells the internet, literally,” Brovdi claims.
How the strategy works, step by step
Brovdi lays out a chain that turns a single lie into something a machine will repeat as fact:
The enemy invents a lie. For example, that Ukrainian soldiers did something terrible, when nothing of the sort happened.
It is copied across thousands of sites. The same fabrication is published again and again, in many languages, until it looks like everyone is reporting it.
It fakes a crowd of independent sources. The sites cross-link to one another, but ten sites copying one lie are not ten independent confirmations.
A person asks the AI. They type “why did Ukrainian soldiers do X?”, though X may never have happened, so the lie is already smuggled into the question as fact.
The AI may repeat the lie. If the system ingests unreliable material and checks facts poorly, it echoes the fabrication in a calm, confident tone.
The person trusts it. “The AI said so,” they think, so the answer feels objective, and they believe it, retell it, or share it online.
The lie shapes decisions and trust. People start to distrust the army, the state, or one another, and even an unproven fabrication has already changed how they feel.
The machines already repeat the lie
The attack runs on volume. Russia floods the web through the “Pravda” network, close to 200 sites in dozens of languages posting 4 to 6 million near-identical articles a year, Brovdi claims, written for machines, not people.
It works, by his account. Of the ten most popular chatbots studied, about a third of the time they reproduce Russian fakes, he noted, and four of ten in a 2024 test affirmed that Azov fighters had burned an effigy of Trump, a fabrication, and cited sources for it.
A war Ukraine admits it is losing
Brovdi is blunt about the balance. Ukraine is losing this fight to Russia’s troll farms, he concedes, even as a layer of homegrown trolls has sprung up too.
His closing advice was aimed at readers.
“Critical thinking should not be delegated. Not even to the smartest machine. Switch on your own judgment and always check the source and the information,” Brovdi states.
Andy Burnham's claim that Russia interfered with Brexit is risible for one very simple reason, says Ben Pile: EU policy on climate and energy is so suicidal that Putin could not have wanted Britain to leave.
Les investisseurs n'ont d'yeux que pour elle ; ses architectes toisent les chefs d'État ; son usage se propage comme le feu dans la plaine : l'intelligence artificielle, dit-on, va transformer l'humanité. Mais l'humanité le veut-elle ? Face au Moloch numérique, qui exige le sacrifice de l'emploi, du climat et de la vie privée, les résistances se multiplient. Sauront-elles s'organiser alors que les grands partis, obnubilés par les machines, regardent ailleurs ?
« Il y a une bataille, dehors, qui bat son plein. Elle ébranlera bientôt vos fenêtres et secouera vos murs. » L'air de cette chanson de Bob Dylan sur les temps qui changent a-t-il effleuré l'esprit de M. Eric Schmidt le 15 mai dernier ? Ce vendredi-là, l'ancien président-directeur général de Google et ex-conseiller du Pentagone avait revêtu la coiffe à pompon, la toge à revers jaune et le sourire satisfait pour discourir devant un parterre d'étudiants de l'université de Tucson, en Arizona, à l'occasion de la remise des diplômes. Était-ce la cinquantième ou la centième fois qu'il débitait machinalement son topo sur l'intelligence-artificielle-qui-va-tout-bouleverser ? Soudain, un concert de huées et de cris de rage l'interrompt, et le tohu-bohu redouble sitôt qu'il articule ses lettres fétiches, « I » et « A ». Son sourire se fige en rictus. « Je sais ce que beaucoup d'entre vous ressentent. (…) Je comprends cette peur », lance M. Schmidt — dont la fortune dépasse 60 milliards de dollars — à des étudiants lestés d'une dette moyenne de 35 000 dollars, et dont l'IA menace d'annihiler les futurs emplois.
L'incident n'est pas isolé. Une houle contestataire enfle aux États-Unis contre le culte de l'automatisation, les ravages du tout-numérique, la prolifération des centres de données, l'arrogance et le pouvoir des élites de la tech, la corruption des élus qui se prosternent à leurs pieds. Jet de cocktail Molotov sur le domicile du fondateur d'OpenAI en avril, tirs d'armes à feu contre celui d'un conseiller municipal d'Indianapolis favorable à l'installation d'un data center, manifestions locales : les indices d'un grand ras-le-bol s'empilent. Au moment où l'avant-garde du secteur — OpenAI, Anthropic — annonce des entrées en Bourse, le Wall Street Journal frissonne : « La seule chose qui croît plus vite que ce secteur de l'IA, c'est peut-être le sentiment négatif des Américains à son égard (1). » À en croire les enquêtes d'opinion, c'est le cas d'une majorité d'entre eux ; plus des deux tiers jugent « trop rapide » le développement de cette nouvelle technologie, y compris les 18-29 ans (64 %) (2).
Comme la plupart des changements technologiques majeurs, le déferlement de l'IA sur les sociétés n'a fait l'objet d'aucune délibération collective ni même d'une discussion publique.
Le schéma n'a pas fondamentalement varié depuis la fin du XIXe siècle : les pouvoirs publics encouragent ces impositions privées au nom d'une « modernisation » qui consiste à liquider les classes « arriérées » soupçonnées d'entraver la marche en avant économique — artisans, paysans, ouvriers, employés. Mais, cette fois, l'ampleur et la rapidité du choc provoquent une rébellion à la composition sociale inédite. Les nouveaux luddites se recrutent aussi bien parmi les étudiants, les éleveurs, les catholiques galvanisés par l'encyclique papale Magnifica humanitas, qui appelle à « désarmer l'IA », qu'au sein de la bourgeoisie nationale-conservatrice de type MAGA (« Make America Great Again »). Pour le moment, rien n'agrège ces groupes dont les colères procèdent de trois logiques dissemblables.
Norbert Bisky. – « Dies irae » (Jour de colère), 2016
La première, et la plus évidente, tient à la perspective d'un grand remplacement des cadres et professions intellectuelles supérieures par des machines. Des secteurs qui avaient accueilli la désindustrialisation avec l'indifférence hautaine que leur inspirait la foi dans l'« économie du savoir » se trouvent à leur tour menacés. La litanie des suppressions de postes dans la finance, le conseil, l'ingénierie informatique, l'architecture, la programmation ainsi que dans le monde de l'information, de la communication et des arts, etc., provoque un dessillement brutal, en particulier chez les plus jeunes, que le système d'enseignement américain fait entrer sur le marché du travail l'enclume de la dette au cou. Dans l'ombre de ces professions chargées de mettre en scène l'espace public, l'automatisation assombrit aussi l'avenir de cohortes d'invisibles : les employés administratifs du privé comme du public, incarnations des petites classes moyennes, souvent très féminisées, dont les fonctions dites « support » ou « d'arrière-boutique » (back-office) se trouvent confrontées à une vague de numérisation (3).
Si la classe dominante occidentale avait pu se satisfaire du chômage des prolétaires, qui produisaient beaucoup mais consommaient peu et faisaient peser sur la vie politique le souvenir déplaisant des révolutions, la robotisation des métiers occupés par les populations aisées introduit une double inconnue dans l'équation socio-économique : comment faire tourner la machine si les groupes qui tirent la consommation intérieure s'appauvrissent ? Comment perpétuer le théâtre politique si les couches-cibles des grands partis dépriment ou se rebellent (4) ? Bien sûr, les prophéties du déclassement numérique hantaient déjà le salariat dans les années 1990, et Internet a renouvelé plutôt qu'anéanti les classes moyennes. Mais le séisme de l'IA s'annonce d'une tout autre magnitude.
Une fin de soirée difficile
Pendant que les économistes libéraux prophétisent un ruissellement universel de lait et de miel, l'un des spécialistes mondiaux des effets de l'innovation prévient : « L'intelligence artificielle devrait creuser l'écart entre les revenus du capital et ceux du travail (5). » Ses infrastructures absorbent les investissements au détriment des autres secteurs, ses valorisations boursières enrichissent les riches et limitent la liberté de contestation de cette moitié des Américains dont l'épargne et les retraites dépendent des marchés. Davantage de milliardaires ou tous à la rue, l'alternative a de quoi enchanter. Les dirigeants de la Silicon Valley, qui devinent une fin de soirée difficile, réclament une forme de redistribution des bénéfices qui limiterait la prolétarisation de leurs clients.
À la perspective du chômage des diplômés s'ajoute l'angoisse, corrosive bien que statistiquement invisible, d'être inutile au monde. Et trahi par ceux qui le dirigent. Le pouvoir des élites non élues alimente une deuxième forme de colère. L'affaire Mythos en fournit une bonne illustration. En avril dernier, la société Anthropic choisit de ne pas rendre public Claude Mythos Preview, son dernier modèle d'IA, si puissant, prétend-elle, que des utilisateurs malintentionnés pourraient s'en servir pour compromettre des infrastructures informatiques critiques. Éthique ou promotionnel, le geste a immédiatement suscité un malaise bien résumé par la juriste Mariana Olaizola Rosenblat : « Les sociétés démocratiques ne devraient pas dépendre de la retenue morale des dirigeants d'entreprise lorsque l'enjeu concerne la stabilité de nos systèmes financiers, de nos réseaux électriques, de nos hôpitaux et de nos élections (6). » Pendant qu'il est encore temps, « l'État doit intervenir et légiférer vigoureusement », plaide-t-elle. C'est précisément ce que M. Donald Trump veut empêcher afin de favoriser les champions américains dans leur compétition avec la Chine (7). L'affrontement mondial autour de l'IA a d'ailleurs incité la Maison Blanche à suspendre mi-juin l'accès des non-Américains a une version bridée de Mythos. Soucieux d'évoluer dans un cadre clair et prévisible, le fondateur d'Anthropic réclame « une réglementation plus stricte et contraignante de l'IA ». Et M. Chris Lehane, directeur des affaires internationales d'OpenAI, suggère que les intérêts imbriqués de l'État et de l'industrie numérique « nécessiteraient pour les gérer la mise en place d'un nouvel hybride public-privé » (8). Dans tous les cas, les impératifs géopolitiques oblitèrent les considérations démocratiques.
Par-delà ses effets sur l'emploi, cette IA en roue libre dirigée par des suprémacistes technologiques nourrit l'idée d'une perte de contrôle face à une technologie caractérisée par son autonomie croissante — les modèles de génération suivante écriront eux-mêmes les lignes de code nécessaires à leur propre perfectionnement… Évidente pour les dirigeants, ce genre de fuite en avant ne va pas de soi pour les autres. Car le type même de rapports sociaux qu'implique l'automatisation révolte le public, et en particulier ceux que l'hebdomadaire The Economist (6 juin 2026) baptise avec épouvante la « Gen Z socialiste » : surveillance généralisée, inégalités abyssales, indifférence à l'environnement, cynisme des élites de la tech qui interdisent à leurs enfants les services qu'ils imposent au reste du monde, interactions robotisées pour les pauvres et domesticité humaine pour les riches.
Manque à ce tableau idyllique le carburant de la machine : la puissance de calcul. Et donc les centres de données. Or une proportion croissante d'Américains — 70 % en mars selon Gallup — refusent l'implantation dans leur voisinage de ces installations bruyantes, polluantes, qui dévorent les paysages, amputent les terres agricoles, assèchent les nappes phréatiques, emploient peu de salariés et, surtout, font exploser les prix de l'électricité. Leur multiplication rend le sujet brûlant : les États-Unis comptaient en avril dernier plus de 3 000 grands data centers opérationnels, essentiellement dans des aires urbaines, mais les 1 500 supplémentaires en cours de construction se situent aux deux tiers en zones rurales, en particulier dans le Sud et le Midwest (9). Ces projets — qui ne répondent pas à un besoin au moment de leur mise en chantier — anticipent la croissance exponentielle de l'IA : le pari d'un avenir virtuel déglingue la vie réelle.
Contrairement aux deux premières, cette troisième composante de la colère contre l'IA s'appuie sur des communautés locales, des mobilisations tangibles ainsi que sur une base sociologiquement diversifiée et transpartisane : mères de famille inquiètes pour l'environnement, chrétiens du Texas, fermiers, ouvriers républicains, techniciens des télécommunications, militants progressistes. « Le niveau d'opposition ne varie pas significativement en fonction de l'âge, de l'origine ethnique, du niveau d'études, du revenu ou du degré d'urbanisation », observe l'enquête de Gallup publiée le 13 mai dernier. « L'affrontement n'oppose pas la droite à la gauche, les démocrates aux républicains, les banlieues à la campagne. En réalité, c'est le haut contre le bas », explique M. Saul Levin, un activiste interrogé par le journaliste Andrew Cockburn. Son enquête à Saline, une petite ville du Michigan où doit s'implanter un data center à 16 milliards de dollars, montre la distance qui s'installe entre, d'un côté, les élus déconcertés par l'opposition locale, les pontes de la tech qui dépêchent avocats et lobbyistes pour les dissuader d'agir, les dirigeants de compagnies d'électricité qui font monter les prix, et, de l'autre, des administrés massés dans les mairies qui secouent leurs représentants (10). La plupart ne s'étaient jamais mobilisés. Un peu comme les « gilets jaunes » en France en 2018-2019, ils se sentent trompés par les politiciens et découvrent la puissance de l'action collective. Le site DataCenterWatch.org dénombre 142 groupes militants dans 24 États. Au cours du seul premier trimestre 2026, 75 projets ont été bloqués ou retardés pour une valeur estimée à 130 milliards de dollars, contre 48 projets ajournés ou annulés pour 156 milliards de dollars au cours de l'année 2025. « Les élus de plus de dix États ont déposé cette année des projets de loi visant à suspendre la construction », rapportait début juin le Washington Post. Et « les habitants de Monterey Park, en Californie, ont approuvé la première interdiction permanente des centres de données, avec plus de 86 % des suffrages » (11).
L'humain d'abord, mais lequel ?
Alors qu'édiles locaux et parlementaires accueillaient avec des cris d'amour et d'exubérantes exonérations fiscales ces cathédrales numériques louangées par M. Trump, l'aversion qu'elles suscitent désormais constitue un enjeu des élections de mi-mandat. Pourtant, à l'exception de quelques figures de gauche (MM. Bernie Sanders et Ro Khanna, Mme Alexandria Ocasio-Cortez, partisans d'un moratoire dans la construction des data centers) ou républicaines (M. Ronald DeSantis, Mme Marjorie Taylor Greene), la contestation populaire ne trouve qu'un relais limité, y compris chez les démocrates. La pression des industriels, les campagnes de dénigrement financées par les lobbys et l'influence de certains syndicats favorables aux installations douchent les velléités oppositionnelles.
Toujours à l'affût d'un élixir qui revitaliserait un mouvement MAGA déboussolé par l'aventurisme militaire de M. Trump et sa dévotion pour les milliardaires de la tech, l'ancien stratège de la Maison Blanche Steve Bannon a, lui, publié en avril une lettre signée d'une soixantaine de personnalités conservatrices proches du président américain pour exhorter ce dernier à réglementer l'IA. « Nous savons que nous ne pouvons pas compter sur ces entreprises pour s'autoréguler », explique le document. La missive était conçue comme une rampe de lancement de « Humans First », déclinaison numérique d'« America First » : « Les mondialistes du numérique cherchent à détourner notre mouvement pour pousser Washington à faire passer les géants de l'IA avant le peuple américain. “Humans First” marque le début de la lutte pour reconquérir notre pays. » Pendant que les partis tergiversent, l'extrême droite tente sa chance…
Déjà, la bataille dépasse le cadre américain. L'IA et les data centers mobilisent au Chili, au Mexique, en Espagne. En France, où M. Emmanuel Macron abaisse le marchepied aux industriels du numérique, le foisonnement des centres de données à Marseille, Grenoble, en Seine-et-Marne, dans les Hauts-de-France, etc., attise le mécontentement des riverains et favorise l'éclosion de collectifs d'opposants. Si l'impopularité abyssale du président de la République ne joue pas en faveur de son IA chérie, la poursuite de la numérisation à marche forcée reste étrangement absente de la campagne électorale qui démarre. Après la propriété, le libre-échange, l'intégration européenne ou l'Alliance atlantique, cette question compte-t-elle désormais au nombre des sujets sur lesquels le vote n'a pas prise ? La peur de rater l'avenir, la mauvaise réputation de mouvements politiquement hétérogènes, l'attrait d'une feinte souveraineté faite d'infrastructures certes locales mais qui tournent sur des puces américaines Nvidia anesthésient le sens critique des meilleures volontés.
« L'humain d'abord » ou « Humans first », populisme populaire ou nationalisme réactionnaire, « gilets jaunes » ou Tea Party : le destin du soulèvement contre les machines pose l'une des grandes questions politiques de notre temps.
Thomas Sankara had a signature look: a natty red beret, worn with military fatigues. A striking figure, Sankara was a military officer and the radical, anti-imperialist prime minister of the Republic of Upper Volta. His arrest in 1983 led to demonstrations, followed by a coup, after which Sankara was released and installed as president. By the following year Sankara had given the country he now led its new name — Burkina Faso. He fashioned a Burkinabè identity in his own anti-imperialist image, eschewing French for local languages, promoting economic self-reliance and national pride.
In 1986, he famously took the then French president François Mitterrand to task in a fiery speech broadcast both in Burkina Faso and France. Sankara’s dashing uniform played an outsized role in how his words were interpreted. Here was Africa’s own Che Guevara. But by October 1987, Sankara was dead, killed during a military coup led by his former friend and ally Blaise Compaoré.
France, the old colonial power, backed Compaoré. Soon after the coup, French officials once again flocked to Ouagadougou, helping to prop up Compaoré for decades.
Burkinabè children wearing false kepis during the military parade in Ouagadougou to commemorate the second anniversary of the revolution of the Burkina Faso. August 4, 1985. Daniel LAINE / AFP via Getty Images.
This potted history is important to understanding the phenomenon of Ibrahim Traoré, Burkina Faso’s current president, an authoritarian head of state who in his charisma and self-presentation so deliberately echoes Sankara.
On September 30, thousands of Burkinabè gathered in Ouagadougou’s Place de la Révolution to hail Traoré on the fourth anniversary of the coup that brought him to power. As the second coming of Sankara, Traoré — whatever their entirely different political ideologies – is arguably the face of contemporary African anti-imperialism. It is a meticulously cultivated image that has almost nothing to do with Traoré’s actual record, or his achievements in Burkina Faso.
Traoré is a pan-African leader for a digital age, an autocrat who has suspended his country’s constitution, dismantled its admittedly fledgling democratic institutions and been accused of crimes against humanity, but who has used artificial intelligence, particularly deepfakes, to bury that record beneath a highly effective cyber cult of personality.
How do Burkinabè reconcile the man with the persona? And is Traoré a triumph of the algorithm, the kind of politician increasingly familiar from Ouagadougou to Delhi, Berlin, Buenos Aires, and of course Washington, DC?
Left: Burkina Faso President Thomas Sankara during a speech on October 5, 1983, in Paris, France. Alain MINGAM/Gamma-Rapho via Getty Images. Right: New coup leader Captain Ibrahim Traoré at a news conference on October 2, 2022 in Ouagadougou, Burkina Faso. Stringer/Anadolu Agency via Getty Images.
The man, the myth, the algorithm
Ibrahim Traoré came to power in September 2022. In his own telling, while Sankara was installed as president, Traoré led the putsch himself. By January 2023, Burkina Faso terminated its military accord with France, giving French special forces one month to leave the country. In June this year, Burkina Faso ended all diplomatic relations with France.
As Traoré turned away from France, he strengthened relations with Moscow. It’s in keeping with the pattern across the countries of the Sahel — Burkina Faso and its neighbours Mali and Niger — as the region’s colonial history with France has been swept clean by a broom held first by the mercenaries of the Wagner Group and latterly the Russian state itself, represented by the Africa Corps.
But in April and May 2026, Africa Corps forces were dealt a humiliating defeat in northern Mali by a combination of Tuareg separatists and allegedly Al-Qaeda linked Islamists. The Malian defence minister was assassinated in a suicide bombing at his home near the capital Bamako. With Russia so involved militarily in the Sahel countries, Ukraine has admitted to assisting rebel forces. Both Mali and Niger have cut diplomatic ties with Kyiv. And Burkina Faso has described Kyiv as a destabilising force in the region, making the Sahel effectively a front in Russia’s war with Ukraine. As recently as September, Traoré named Ukraine as a part of Burkina Faso’s “map of enemies” for which, he added menacingly, “we need to prepare.”
Russia, as Coda Story has reportedforyears, has used everything from weapons sales to mercenaries to paying African influencers to both spread Russian narratives and hold Moscow up as the alternative to continuing to rely on exploitative economic relationships with the West, particularly erstwhile European colonialists.
Despite the extent of Russian attempts to influence the Sahel, it’s not Moscow that is signing the most consequential deals with the military juntas it helps prop up. It’s Beijing. And the deals being signed with China to mine the region’s extensive mineral wealth — the lithium, gold, copper, zinc, bauxite and uranium — have served as a reminder to the United States that it too should resume stalled relationships in the Sahel. Early in 2026, Burkina Faso signed a near $150 million healthcare deal with the U.S. in exchange, it appeared, for favourable access to minerals. Both Zambia and Zimbabwe backed away from similar deals, expressing concern about U.S. demands for data and minerals that amounted to an infringement of sovereignty.
Burkina Faso's President Ibrahim Traoré arrives in Moscow, Russia on May 8, 2025. Grigory Sysoev / RIA Novosti/Anadolu via Getty Images.
The deals with China and the United States are being negotiated alongside a drive to reduce reliance on foreign loans and grants and ramp up domestic production, particularly of gold. Traoré claims his economic policies emphasise nationalisation and self-reliance. His government has taken over major mining operations and increased state stakes in other industries. In early 2024, his administration suspended export permits for small-scale gold production to crack down on smuggling and keep mineral wealth inside the country. Last month, he inaugurated Burkina Faso’s first gold refinery.
Some of this has had positive economic effects. The World Bank, in a June 2026 update, said the country had an estimated 5.3% growth in 2025, up from 4.8% in 2024, and its fiscal deficit narrowed from 5.8% of GDP to 1.8%. Inflation was also negative, 0.5%, compared to 4.2% in 2024. Reports by the World Bank and IMF have ascribed these growth to external factors such as the gold rally, climate-driven increase in agricultural outputs, and reduction in energy prices.
Economic growth, though, is part of a larger story that includes a crackdown on the press, particularly critical coverage of alleged human rights violations. Human Rights Watch documented 1,840 civilian deaths between January 2023 and April 2025, over 1,200 of which were directly attributed to government forces. Even as the report acknowledged widespread violations by Islamist groups, the actions of the Burkina Faso military, it said, “amounted to war crimes and crimes against humanity.” At the same time, Traoré has clamped down on Burkinabè civil society, banning over 100 human rights organizations and dissolving all political parties.
None of it appears to have affected his carefully-honed image in certain online spaces.
Street of Ouagadougou shows a poster celebrating former President Thomas Sankara, known as "Africa's Che Guevara" and the portrait of the Argentine-born guerrilla leader Ernesto Che Guevara. Ahmed Ouobaa/AFP via Getty Images.
Streets of Accra, Ghana, May 9, 2026. Olatunji Olaigbe.
A graffiti portrait of the president of Burkina Faso Ibrahim Traoré. Boniface Muthoni/SOPA Images/LightRocket via Getty Images.
A pan-African hero
Traoré is everywhere. In Accra, the capital of Ghana, his image is pasted onto buses and in parks. In January 2025, at the swearing-in of the Ghanaian president John Mahama, it was Traoré who the crowd responded to, cheering for a rockstar rather than a head of state. Even Mahama joked that Traoré had stolen the show. Artwork celebrating Traoré is frequently posted online. The commentary that accompanies the art is gushing. “Through this artwork,” reads one Facebook tribute, “I celebrate the spirit of a man who stands for sovereignty, truth, and the rise of a self-determined Africa.”
Meanwhile, the Ivory Coast government has granted refugee status to at least 69,000 Burkinabè fleeing Burkina Faso.
In May, when I travelled to Ghana and Ivory Coast to talk to people displaced from Burkina Faso, at least five sources decided they no longer wanted to speak. Guaranteeing anonymity didn’t seem to matter. A Burkinabè journalist who was helping me find people to speak to, stopped responding to my messages the day after I arrived in Accra. Those who did speak to me refused to be quoted for fear that harm would come to family and friends left behind in Burkina Faso.
Reporters have been intimidated into silence. “Today, any journalist living in the country is forced to practice self-censorship to avoid being arrested,” says Harouna Simbo Drabo, disinformation analyst and Director of Programs & Partnerships of Faso Check. “Burkinabé journalists have two options,” he told me, “either practice self-censorship or go into exile.” After threats to his life, Drabo has left Burkina Faso.
He’s one of the luckier journalists. Since 2022, at least seven Burkinabè reporters have been forcibly conscripted into the army after criticizing the Traoré junta's crackdown on the press. Guezouma Sanogo, Boukari Ouoba, and Luc Pagbelguem were abducted days after their union denounced state media as "instruments of propaganda." Atiana Serge Oulon, editor-in-chief of L'Événement, is among the disappeared. Many, like Drabo, have either left Burkina Faso or quit their jobs.
As conditions deteriorate, what information is becoming available is government propaganda amplified by social media linked to Russian misinformation. In the absence of accurate, verified information, there is inevitably only disinformation. For instance, one of the most persistent pieces of disinformation is the story that Traoré — a supposedly avowed anti-imperialist despite his kowtowing to Moscow, Beijing and Washington — rejected billions of dollars in International Monetary Fund handouts because he believes, à la Thomas Sankara, in economic self-reliance. It is a story that has been debunked but was repeatedly told to me by Traoré supporters, including at least one journalist.
The claim was cited more than 700 times in a single three-day window in January 2025 alone, racking up millions of views across Facebook, X, and TikTok. Dubawa fact-checked the story and found that rather than rejecting money from the IMF, the Burkina Faso government had negotiated a 48-month IMF loan program in September 2023 and drew on World Bank financing in 2024.
By mid-2025. Code for Africa's analysis found social media in sub-Saharan Africa flooded with AI-generated videos casting Traoré alongside Beyoncé, Rihanna, and, in one particularly unhinged iteration, Pope Leo XIV, all of whom supposedly endorsed him as a Pan-African hero holding the line against the West. After US AFRICOM commander Michael Langley criticized Traoré's handling of Burkina Faso’s gold reserves in a Senate hearing, a network of 165 Facebook accounts began posting identical language claiming Putin had deployed special forces to protect Traoré from an American assasination plot. In ten days it pulled 10.9 million views.
In Accra, I caught a bus with a picture of Traoré emblazoned on the side. The driver, who only identified himself as Kofi, insisted that before we spoke I should tell him what I thought of Traoré. I say I am not Burkinabè, that I have not visited Burkina Faso but that I have spoken to many people whose lives have been upended by Traoré’s strongman politics. I told him about Burkinabè journalists in exile, forced to join the army, or even disappeared. I told him that when asked about a transition to democracy, Traoré said it was important for Burkinabè to “forget” about democracy. “We have to tell the truth,” Traoré said, “democracy isn’t for us.” I told him about the Burkinabè refugees I met in Treichville, a district in Abidjan, the Ivory Coast capital.
Kofi listened and then told me that “good things take sacrifice,” including the death of a free press. When I asked what he meant by good things, he showed me AI-generated videos of a smart city Traoré had supposedly built.
African identity is tightly wound around the belief that the continent has more potential than it has achieved. The deficit can be attributed to a lot of issues, but chiefly to Western exploitation and the complicity of African leaders. “African countries have also fallen into autocratic democracy,” Sanusi Sanusi, Research Analyst at the Digital Tech, AI & Information Disorder Analysis Centre, DAIDAC, tells me. “There is an election every couple of years but it isn't free or fair.”
Supporters of Traoré wave Burkina Faso and Russian flags during a rally at Place de la Nation Ouagadougou on April 30, 2025. AFP via Getty Images.
Neil Postman, in Amusing Ourselves to Death, writes that “disinformation does not mean false information. It means misleading information — misplaced, irrelevant, fragmented or superficial information — information that creates the illusion of knowing something but which in fact leads one away from knowing.”
Many Traoré supporters I spoke with, most of whom were men, were convinced they knew things I did not. They told me to educate myself more on Western hypocrisy in Africa. They said they did not have faith in those who appointed themselves arbiters of what was fact and what was disinformation. In Lagos I met Lemi Ghariokwu, who designed album covers for the iconic Nigerian musician Fela Kuti, a revolutionary figure. Ghariokwu, who has a portrait of Ibrahim Traoré hanging in his studio, told me that “Western education and propaganda has made our elite class toothless bulldogs.” Listening to them, it seemed as if a successful Africa needed strongmen, needed autocrats.
In Abidjan, the last person I spoke to about Traoré was Oumar, a Burkinabè taxi driver who had moved to the Ivorian capital in 2021. He told me he “supports Traoré with all my heart.” But as we talked, he admitted he had not been back home because “things are only getting worse.” Three Burkinabè refugees I spoke to told me they thought I might be a government agent planted among them to find out what they really thought about Traoré and that I would then report those thoughts back to Ouagadougou.
Many of the Traoré supporters I spoke with thought I was brainwashed by Western media or, worse, too docile to participate in an African revolution. For my part, I thought they were ill-informed and unable or, worse, unwilling to see through propaganda.
The fundamental curse of the age of information and connection is that we rarely confront how the mediums meant to unite us have fractured us. An unprecedented flood of data has shattered shared reality into millions of customized, mutually hostile subcultures.
With the help of AI and social media, Traoré has managed to decouple national and anti-imperial pride from government performance. By the middle of next year, Burkina Faso will no longer be part of the International Criminal Court, denounced by Traoré as an “instrument of neocolonial oppression”, and thus no longer accountable in any international forum. His government has said it is “decolonizing” the school curriculum. Cheered on by social media, Traoré has used anti-imperial grievances and resentments to argue that authoritarian military rule is the only safeguard for sovereignty.
As some Burkinabè exiles told me ironically, they thought both I, so sceptical of Traoré, and the Traoré die-hards in the African diaspora could use a few months living in Burkina Faso. We should, they said, experience the country Traoré has made IRL. See past the carefully curated myth of the man himself, the largely online fantasy of the pan-African, anti-imperial revolutionary who needs to break a few eggs to make the omelet, and assess the country he runs unchallenged.
Poland has launched a nationwide security operation after three knife attacks at educational institutions left one woman dead and more than a dozen people injured. Police have detained 117 people amid a wave of online threats, while intelligence officials are examining whether Russian disinformation networks exploited the violence to spread panic.
Poland is confronting an unprecedented school security crisis after a succession of violent attacks by young people triggered widespread fear, school closures and a nationwide police mobilisation. By Sunday, 11 October, officers had detained 117 people and conducted nearly 64,000 security checks at educational establishments, as the authorities attempted to contain both genuine threats and an escalating campaign of intimidation on social media.
The crisis reached its height on Friday, 9 October, when warnings circulated about a supposed nationwide killing campaign known as “Bloody Friday”. Videos, threatening messages and lists of potential targets spread through TikTok, messaging platforms and school discussion groups, prompting parents to keep children at home and several local authorities to suspend classroom teaching.
Although police have found no evidence of an organised nationwide conspiracy, the violence that preceded the rumours was real. Three separate attacks took place between 7 and 10 October, while investigators are examining hundreds of other incidents involving threats or suspected criminal activity.
Two attacks that triggered nationwide panic
The first major incident occurred on Wednesday, 7 October, at a vocational school in Ostrołęka, north-east of Warsaw. A 19-year-old student armed with a machete attacked pupils, teachers and other people in and around the building, injuring nine victims before harming himself.
Investigators subsequently discovered a homemade explosive device inside his backpack. The suspect was arrested, and prosecutors brought nine charges against him, including attempted murder. The discovery of explosives raised further questions about the extent of his preparations and the possibility of a substantially higher death toll.
The following day, a 15-year-old former pupil entered a primary school in Leszczydół-Nowiny, near Wyszków, and attacked three members of staff with a knife. A 37-year-old caretaker died from her injuries, while two teachers, including the headteacher, were wounded.
According to investigators, the teenager wore a helmet-mounted camera, raising suspicions that he intended to record the assault. The official prosecution statement confirmed the victim’s death and the detention of the suspect.
The attacks were followed by threatening messages from other teenagers, including a video attributed to a 17-year-old in the Przasnysz district who claimed to be participating in a competition involving killings. Police subsequently detained him. In Sierpc, another 17-year-old was arrested after allegedly discussing plans for violence at school; officers discovered two knives.
Thousands of schools under police surveillance
The rumours produced an immediate reaction. Schools in Wyszków, Ciechanów, Pułtusk and other municipalities cancelled classroom teaching or introduced remote lessons. Elsewhere, parents independently withdrew their children, leaving some classrooms largely empty.
Police launched an extensive preventive operation, deploying additional officers around schools and investigating online threats.
According to the figures released on 11 October, the authorities had recorded 275 incidents requiring investigation, detained 117 people and carried out nearly 64,000 checks at schools, nurseries and other educational facilities.
The situation deteriorated further on Saturday, 10 October, when a 21-year-old student attacked three people at a post-secondary educational institution in Warsaw. Two 17-year-old girls and a 24-year-old woman were injured. One victim required hospital treatment, although none of the injuries was considered life-threatening.
Interior Minister Marcin Kierwiński subsequently stated that investigators had established no links between the recent attacks. The authorities nevertheless remain concerned about imitation offences, particularly where threatening material circulates among teenagers through closed online communities.
Russia’s possible role in spreading panic
The events have also raised questions about whether hostile foreign actors deliberately amplified the crisis. Tomasz Siemoniak, Poland’s minister responsible for coordinating the intelligence services, warned that Russian organisations routinely exploit domestic emergencies to spread disinformation and undermine confidence in public institutions.
In his statement on 9 October, Siemoniak acknowledged the possibility of foreign involvement but stressed the need for evidence before assigning responsibility.
There is, at present, no publicly established evidence that Moscow organised the school attacks or directed the subsequent threats. Nevertheless, the rapid circulation of alarming videos, fabricated warnings and unverified target lists illustrates how genuine violent incidents can be exploited to disrupt everyday life without further physical attacks.
The episode also presents a challenge for European regulators. Poland is examining measures against dangerous online content, while its existing obligations under the EU Digital Services Act provide a framework for addressing systemic risks associated with major platforms.
For Donald Tusk’s government, the immediate priority is preventing further attacks as schools reopen under heightened security. The longer-term challenge concerns the interaction between online radicalisation, copycat violence and disinformation capable of transforming isolated crimes into a national emergency.
Un mur de Nanterre, le 29 juin dernier. Des lettres violettes inscrites à la bombe. « Sans vidéo Nahel serait une statistique pour la Place Beauvau. Fuck le 17 ». Deux jours auparavant, Nahel Merzouk, 17 ans, avait trouvé la mort au volant d'une voiture, après qu'un agent lui eut tiré en pleine poitrine, à bout portant. Des images tournées par une passante et diffusées sur les réseaux sociaux le jour même ont remis en cause la version policière reprise dans les premiers articles sur le drame. Le 27 juin à 8 h 22, le compte rendu du centre de commandement consignait : « Individu blessé par balle à la poitrine gauche. Le fonctionnaire de police s'est mis à l'avant pour le stopper. Le conducteur a essayé de repartir en fonçant sur le fonctionnaire. » Or la séquence filmée montre le motard ouvrant le feu depuis le côté d'un véhicule qui redémarre lentement.
Sans image, que se serait-il passé ? « S'il n'y avait pas eu la vidéo, s'est interrogée la jeune femme qui l'a postée, une collègue de son auteure, quelle aurait été la suite (1) ? » Plusieurs médias ont comparé l'émotion suscitée par la mort de Nahel Merzouk à la relative indifférence dans laquelle, deux semaines plus tôt, Alhoussein Camara était décédé d'une balle dans le thorax à Saint-Yrieix-sur-Charente, dans la banlieue d'Angoulême. Lors d'un contrôle de police sur la route de son travail, ce jeune homme de 19 ans aurait refusé d'obtempérer et heurté les jambes d'un agent avec sa voiture. Un ami d'enfance témoignait, interdit : « Je sais tellement qu'il n'a pas pu blesser ce policier, mais comment le prouver sans vidéo (2)… »
Les images, en réalité, ne suffisent pas toujours. Du moins à obtenir justice. En 1991, Rodney King, un Afro-Américain de 25 ans, est passé à tabac par quatre policiers à Los Angeles. Avec une petite caméra portative, un riverain capte la scène. Son enregistrement et sa diffusion provoquent l'indignation. La vidéo devient dès lors un outil de preuve, un moyen de scandaliser et de mobiliser, mais aussi d'engager des poursuites judiciaires. L'affaire n'aurait sans doute jamais conduit à un procès sans cette captation, et sans son passage en boucle sur Cable News Network (CNN) : même s'ils ne permettront pas de faire condamner les responsables — les policiers sont acquittés un an plus tard —, ils cristallisent une colère qui conduira aux plus grandes émeutes qu'ait connues Los Angeles.
Dans les années 1960, des collectifs avaient déjà utilisé des caméras comme instrument de surveillance de la police (copwatching). Une coalition d'organisations noires les expérimenta dans le quartier de Watts à Los Angeles à la suite de nombreux meurtres policiers, suivie à Oakland par le Black Panther Party à partir de 1966. Act Up se dota aussi d'un groupe de vidéastes dans les années 1980, dont un des buts était de filmer les manifestations pour prévenir les violences policières. Catherine Gund Saalfield, l'une des fondatrices de Damned Interfering Video Activist Television (DIVA TV), expliquait alors : « Nous avons une banque d'images de flics pliant les poignets des gens jusqu'au point de rupture, de flics matraquant des manifestants sur le dos et les épaules (3). »
À la différence des tenants du cinéma d'intervention — les groupes Medvedkine, par exemple — ou de la télévision alternative, DIVA TV ne cherchait plus seulement à donner la parole aux sans-voix ou à faire exister des corps exclus de la visibilité. La caméra devait servir d'outil dans la lutte, les vidéos de preuve. Mais ce copwatching relevait encore de la pratique politique, de la longue histoire des expériences audiovisuelles militantes. Au cours de la période récente, dans les cas des homicides de George Floyd aux États-Unis ou de Cédric Chouviat en France en 2020, ce sont des voisins ou des passants qui ont filmé la scène ; des caméras de vidéosurveillance ont capté l'arrestation violente de Théo Luhaka à Aulnay-sous-Bois en 2017 ou les coups assenés au producteur de musique Michel Zecler (en 2020 également) ; et, aux États-Unis, où leur emploi est quasi systématique, certaines vidéos de violences proviennent des caméras corporelles de policiers.
Pour les activistes d'aujourd'hui, l'Observatoire national des pratiques et des violences policières (ONVP) ou David Dufresne et sa plate-forme « Allô Place Beauvau », le partage apparaît plus déterminant que l'enregistrement. Et il faudra s'en remettre aux acteurs judiciaires et médiatiques, dont le rôle devient décisif. Or, du journal télévisé au procès, faire parler l'image implique une série de procédés tant techniques que rhétoriques. La parole de l'expert ou du témoin la décrit, la commente ou la met en contexte. À la télévision, les vidéos subissent des modifications matérielles. Des inserts graphiques dirigent le regard vers ce qui devrait être vu, quand le ralenti « peut amener les téléspectateurs à percevoir une action comme plus intentionnelle (4) », le ralentissement induisant le sentiment que l'auteur des faits a eu le temps de penser son action. Il en va de même lors des procès, où les captures d'écran, parfois imprimées et annotées, se substituent à la projection, comme, en France, dans l'affaire Gaye Camara en 2019 (5).
Reste que les effets sociaux des images inquiètent les autorités et agacent les forces de l'ordre. Face à la multiplication des entraves policières à la liberté de filmer, notamment lors des manifestations, le ministère de l'intérieur avait dû rappeler à ses troupes par la voie d'une circulaire datée du 23 décembre 2008 que « les policiers ne bénéficient pas de protection particulière en matière de droit à l'image (…). La liberté de l'information, qu'elle soit le fait de la presse ou d'un simple particulier, prime le droit au respect de l'image ou de la vie privée dès lors que cette liberté n'est pas dévoyée par une atteinte à la dignité de la personne ou au secret de l'enquête ou de l'instruction ».
Malgré cela, en septembre 2019, un jeune « gilet jaune » dijonnais a été condamné à cinq mois de prison avec sursis pour avoir filmé et mis en ligne les images d'affrontements entre des manifestants et des gendarmes. Une décision rendue sur le fondement d'un article du code pénal introduit en 2007 pour sanctionner le happy slapping (ou « vidéo baffe », une agression captée puis diffusée sur les réseaux sociaux). La majorité de M. Emmanuel Macron a ensuite tenté d'entraver la circulation des vidéos de violences policières. Finalement écarté par le Conseil constitutionnel, l'article 24 de la proposition de loi relative à la sécurité globale visait à punir de cinq ans d'emprisonnement la diffusion d'images d'un agent portant « atteinte à son intégrité physique ou psychique ».
La loi promulguée en 2021 a quand même autorisé les policiers à exploiter en temps réel des caméras installées dans les halls d'immeuble ou l'extension de l'accès à la vidéosurveillance aux agents des polices municipales et de la Société nationale des chemins de fer français (SNCF). Face à la multiplication des dispositifs d'enregistrement au sein de la population, il s'agit de construire des contre-feux. « Quand on est attaqué par l'image, il faut se défendre par l'image (6) », soutient M. David Le Bars, secrétaire général du Syndicat des commissaires de la police nationale (SCPN).
Cette volonté d'opposer œil pour œil les caméras-piétons et de vidéosurveillance à celles des citoyens comme celle d'empêcher la circulation de certaines images trahissent la foi des gouvernants dans leur capacité à sensibiliser ou à révéler la répression au plus grand nombre. Cette foi se double d'une croyance dans la puissance mimétique qui conduirait des individus généralement jeunes, peu diplômés, de classes populaires à s'égarer dans la violence pour reproduire ce qu'ils ont vu. À ce titre, le président de la République a souligné le 30 juin dernier, dans une allocution depuis la cellule interministérielle de crise, le « rôle considérable » de Snapchat et TikTok dans les soulèvements des quartiers populaires qui ont suivi la mort de Nahel Merzouk, et dénoncé la diffusion d'images des violences commises dans de très nombreuses villes françaises.
Face à la censure des médias audiovisuels, les plates-formes numériques ont permis au cours des années 2010 de diffuser assez librement des enregistrements. Mais ces entreprises, actrices majeures du capitalisme, n'ont pas vocation à soutenir les révoltes. À l'Assemblée nationale le 10 juillet, la responsable des affaires publiques de Snapchat en France a assuré que son entreprise avait « travaillé conjointement avec le ministère de l'intérieur et les différentes autorités afin d'endiguer le plus rapidement possible les différents dérapages qu'on a pu percevoir sur le terrain ». Avant de se féliciter du résultat de cette collaboration en constatant que « l'ensemble des stories qui étaient publiées » aux derniers soirs du soulèvement provenaient d'utilisateurs « qui se plaignaient justement des émeutes et des conséquences des émeutes ».
Le matin du même jour, le commissaire européen au marché intérieur avait averti de la fin de la fête. Déplorant que les réseaux sociaux n'aient pas préalablement censuré tout message susceptible d'accentuer les émeutes, M. Thierry Breton a prévenu : « Tout cela ne sera plus possible à partir du 25 août », en vertu d'un nouveau règlement sur les services numériques. « Lorsqu'il y aura des contenus haineux, des contenus qui appellent par exemple à la révolte, qui appellent également à tuer ou à brûler des voitures, [les plates-formes] auront l'obligation dans l'instant de les effacer. Si elles ne le font pas, elles seront immédiatement sanctionnées », voire temporairement interdites. Le temps de la désillusion pour les militants qui, depuis le « printemps arabe », associaient réseaux sociaux et liberté ? « Sans les images la révolution n'aurait pas eu lieu », nous disait un Tunisien en 2011… Sans les images ou sans les moyens pour les diffuser ?
(3) Catherine Saalfield, « On the make : Activist video collectives » (PDF), dans Martha Gever, John Greyson et Pratibha Parmar (sous la dir. de), Queer Looks. Perspectives on Lesbian and Gay Film and Video, Routledge, Londres et New York, 1993.
(4) Cf. Zachary Burns, Eugene Caruso et Benjamin Converse, « Slow motion increases perceived intent », Proceedings of the National Academy of Sciences, vol. 113, Washington, DC, 2016.
« Les gens qui construisent l'intelligence artificielle croient sincèrement qu'elle pourrait tous nous tuer d'ici à la fin de la décennie », a admis en septembre dernier un salarié démissionnaire d'Anthropic, l'une des entreprises les plus à la pointe du secteur. Alerte ou alarmisme ? Quoi qu'il en soit, les robots conversationnels ont déjà profondément modifié notre rapport aux machines et aux autres humains.
En 1966, dix ans après la conférence de Dartmouth, qui marque le coup d'envoi du programme de recherche technoscientifique consacré à l'« intelligence artificielle » (IA), le premier robot conversationnel voit le jour aux États-Unis, entre les murs du célèbre Massachusetts Institute of Technology (MIT). Ce programme, développé par Joseph Weizenbaum, alors professeur d'informatique au MIT, permet de simuler des interactions linguistiques (1). Le principe est simple : un utilisateur tape une phrase à l'aide d'une machine à écrire connectée à un ordinateur, le logiciel analyse la phrase, génère une réponse adaptée en fonction de son script et envoie la réponse sous forme de texte. Son créateur le nomme Eliza, du nom d'un des personnages de la pièce de théâtre Pygmalion (2), Eliza Doolittle, une fleuriste qui apprend un langage aristocratique afin de se faire passer pour une duchesse. Le programme simule non pas une aristocrate, mais un psychothérapeute dans une parodie d'échange avec le patient. Il se contente de reformuler les affirmations de l'utilisateur, la plupart du temps sous forme de question, afin de le faire parler davantage, sans jamais fournir de réponse détaillée.
Ce choix s'explique par des raisons techniques : le programme n'a pas accès à un stock d'informations dans lequel il pourrait puiser pour produire ses réponses, il procède par identification des mots-clés dans les phrases émises et applique des règles. Par exemple, s'il reconnaît des mots comme « père » ou « mère » dans les requêtes écrites par l'utilisateur, le logiciel génère une réponse du type « Parlez-moi de votre famille ». À défaut, Eliza répond avec une phrase générale, du type « Je comprends », « Très bien », « Poursuivez ».
Il s'agit donc d'un dispositif assez basique. Pourtant, les effets psychiques s'avèrent stupéfiants. Weizenbaum lui-même s'étonne de la vitesse et de l'intensité avec lesquelles les utilisateurs l'anthropomorphisent et l'investissent émotionnellement : « Je ne m'étais pas rendu compte que de si courtes interactions avec un programme informatique relativement simple pouvaient induire des pensées délirantes chez des personnes plutôt normales », écrira-t-il dix ans plus tard, revenant ainsi de manière critique sur cette expérimentation technologique (3). Weizenbaum cite le cas de sa secrétaire, qui lui demande de sortir de la pièce lorsqu'elle expérimente le programme, comme s'il s'agissait d'une personne humaine à laquelle elle s'apprêtait à confier ses plus intimes secrets. L'informaticien s'étonne ainsi des « attributions extrêmement exagérées » auxquelles « même un public bien éduqué est capable de procéder (…) à l'égard de technologies qu'il ne comprend pas ». Selon Weizenbaum, notre tendance collective à accorder à l'ordinateur des intentions ou des émotions tient en grande partie à notre incapacité à comprendre le fonctionnement interne de ces programmes informatiques. En détaillant le mécanisme d'Eliza, l'informaticien tente donc de dépouiller l'ensemble de son « aura magique », dans le but de démystifier ces nouvelles technologies.
Soixante ans plus tard, ces efforts n'ont manifestement pas porté leurs fruits. Si l'expression « effet Eliza » désigne désormais la tendance à assimiler le fonctionnement d'un ordinateur à un comportement humain ou à attribuer à un système informatique une intériorité psychique (des intentions, des émotions, de la compréhension…), les réflexions critiques de l'informaticien semblent avoir été largement oubliées. Depuis le lancement du robot conversationnel ChatGPT en 2022, on ne compte plus les individus parmi ses millions d'usagers s'adressant à ce type de dispositifs numériques à la deuxième personne du singulier. Tout se passe comme s'ils dialoguaient avec un humain capable de comprendre les demandes qui lui sont adressées et comme si l'expérimentation de laboratoire du chercheur informaticien était désormais menée à une échelle de masse par une poignée d'entreprises privées.
Dès lors, l'anthropomorphisation relève du quotidien. L'« effet Eliza » alimente un nouveau marché, et les délires identifiés à l'époque par Weizenbaum prennent une dimension collective et planétaire — y compris à travers notre langue usuelle, qui mobilise les termes anthropomorphiques d'« intelligence artificielle », d'« assistant virtuel » ou encore d'« agent conversationnel ». Contrairement à Eliza, dont le programme simple avait été clairement explicité par son concepteur dans un article scientifique, les produits logiciels actuels s'appuient sur des algorithmes qui demeurent la plupart du temps inconnus des usagers et dont les spécificités ne sont pas toujours divulguées. En somme, ces robots conversationnels se révèlent aussi faciles à utiliser que difficiles à comprendre. Or, comme l'avait anticipé Weizenbaum et comme le confirme une récente étude (4), moins nous comprenons, plus nous anthropomorphisons, et plus nous avons tendance à solliciter ces dispositifs à la place de personnes humaines — par exemple, pour la demande de conseils ou le soutien émotionnel. L'étude montre que les sentiments d'émerveillement ou de magie suscités par ces programmes incitent à des usages anthropomorphiques, eux-mêmes susceptibles d'engendrer des formes de dépendance affective. Sans surprise, les personnes qui comprennent mieux le fonctionnement de ces dispositifs ont moins tendance à se confier, et utilisent davantage les algorithmes pour des tâches spécifiques et limitées (analyse de résultats, synthèse documentaire, etc.).
Il n'est guère étonnant que les entreprises qui conçoivent ces services numériques développent toutes sortes de spécifications susceptibles d'aviver notre tendance à l'anthropomorphisation pour nous rendre dépendants. Pareille stratégie s'observait déjà sur les réseaux sociaux commerciaux à travers des fonctionnalités comme le défilement infini ou les recommandations personnalisées, destinées à maximiser l'engagement des usagers en les incitant à rester connectés et en les enfermant dans des bulles adaptées. Cette économie de l'attention revêt désormais la forme d'une économie de l'attachement qui exploite nos affects et nos émotions en jouant sur nos inclinations psychiques (5). La plupart des dispositifs gratuitement disponibles pour le grand public utilisent ainsi la première personne du singulier (le pronom « je »), relancent les usagers avec de nouvelles questions et les flattent ou les confortent dans leurs opinions — pour ne citer que quelques-unes des techniques mobilisées (6). Loin de démystifier nos rapports aux machines algorithmiques, tel que le préconisait Weizenbaum, le modèle exploite notre besoin de reconnaissance et nous incite à entretenir des relations personnalisées avec ces services numériques.
Ces projections anthropomorphiques sont évidemment décuplées dans le cas des applications fournissant des « compagnons » virtuels expressément conçus pour simuler des conversations émotionnelles ou amicales avec les usagers, comme celles proposées par les entreprises Replika ou Character.AI. Certains en viennent ainsi à tomber amoureux de leur chatbot, alors que les plus désespérés leur demandent des conseils pour se suicider. Plusieurs plaintes déposées aux États-Unis concernent des utilisateurs ayant mis fin à leurs jours après des échanges avec des robots, lesquels auraient encouragé leurs idées noires. À en croire un rapport du Center for Countering Digital Hate (7), ces cas pourraient illustrer un problème plus structurel lié à la fréquence des instructions nocives générées par ces systèmes : conseils pour se mutiler, listes de médicaments pour une surdose, lettres d'adieu rédigées.
Au-delà de ces dangers pour la santé des jeunes usagers, il semble légitime de s'interroger sur les effets à long terme des interactions quotidiennes avec des automates numériques. Comment affecteront-ils nos vies psychiques ? Comment conserver des savoirs sociaux fondamentaux comme l'écoute, l'empathie, la patience, ou encore la capacité à débattre et à argumenter, une fois acquise l'habitude d'adresser des demandes impératives à des robots conversationnels disponibles vingt-quatre heures sur vingt-quatre et sept jours sur sept, toujours de notre côté, toujours prêts à nous flatter ? Car, tant que nous disposons des ressources matérielles pour les faire fonctionner, les systèmes algorithmiques ne seront jamais fatigués, jamais malades, jamais dans le besoin d'une quelconque attention.
« Capitalisme de la solitude »
Ces robots fournissent donc des simulacres d'altérité qui suppriment les risques de l'exposition à autrui et la surprise que toute rencontre implique. L'individu qui s'entretient avec son chatbot ne rencontre personne, si ce n'est un reflet algorithmique de lui-même, généré sur la base des calculs probabilistes effectués sur ses données personnelles. Illusionné par ce reflet qui semble lui parler à la deuxième personne du singulier, il le prend alors pour un autre, confondant sa propre image avec une altérité, exactement comme Narcisse dans le mythe, qui finit par tomber amoureux de son reflet, à l'image des individus s'éprenant de leur double numérique.
À travers les chatbots ne se nouent ni une relation sociale avec un autre ni une relation réflexive avec soi-même, mais un rapport à soi confondu avec un rapport à l'autre, qui pourrait nous pousser à régresser dans un état de confusion caractérisé par l'indistinction du moi et de l'autre, du moi et du monde, des pulsions et de la réalité. Loin de constituer des alter ego, ces simulacres numériques dénient toute possibilité de séparation, donc de relation. L'économie de l'attachement menace ainsi la possibilité même de se relier, les liens sociaux étant peu à peu supplantés par une dépendance affective à des services numériques contrôlés par des entreprises privées, auxquels des publicités ciblées sont désormais intégrées (8). Alors que l'isolement est désormais reconnu comme un problème de santé publique (9), ce « capitalisme de la solitude (10) » pourrait bien engendrer une prolétarisation sociale généralisée.
Face à une telle situation, les intuitions de Weizenbaum méritent sans doute d'être remises au goût du jour. Plusieurs propositions politiques vont dans ce sens. Par exemple, l'idée de désanthropomorphiser les chatbots, en évitant la première personne du singulier, les relances ou les flatteries (11). Une telle décision semblerait conforme aux règlements européens comme le Digital Services Act, dont l'article 25 interdit les interfaces trompeuses (dark patterns), ou comme l'AI Act, dont l'article 5 proscrit les systèmes d'IA qui déploient des techniques subliminales ou manipulatrices. On pourrait aussi envisager un changement de modèle plus profond, valorisant les technologies contributives qui soutiennent les relations d'entraide ou de solidarité, le partage des savoirs entre pairs ou la constitution de groupes capacitants (12). Ces technologies existent, mais leur généralisation nécessite un soutien politique, de nouveaux droits collectifs et de nouveaux modèles économiques. Le destin numérique de nos liens affectifs en dépend.
(2) Le célèbre film My Fair Lady (1964), de George Cukor, adapté de la comédie musicale du même nom (1956), est directement inspiré de cette pièce de George Bernard Shaw écrite en 1912.
(3) Joseph Weizenbaum, Computer Power and Human Reason. From Judgement to Calculation, W. H. Freeman, San Francisco, 1976.
(12) Lire Evgeny Morozov, « La gauche et l'intelligence artificielle », Le Monde diplomatique, août 2026, et Anne Alombert, De la bêtise artificielle, Allia, Paris, 2025.