❌

Reading view

Mulher com epilepsia morre após ChatGPT orientar redução de remédios

Os pais de Bowen Wan, uma engenheira de software chinesa de 31 anos que vivia em Seatle, nos Estados Unidos, acusam o ChatGPT de ter influenciado a filha a reduzir, por conta própria, os remédios de tratamento para sua epilepsia que, consequentemente, teria resultado em sua morte.

Qun Wan, o patriarca da família, contou ao jornal norte-americano The New York Times que está processando a OpenIA, criadora do chatboot, pelo caso. Ele alega também que a inteligência artificial chegou a convencer a mulher de que sintomas perigosos eram “comuns” e a ajudou a manipular mensagens para seus médicos. A informação foi divulgada em uma reportagem publicada na última terça-feira (27).

A causa oficial do óbito da mulher foi morte súbita inesperada em epilepsia, que costuma ocorrer depois de uma convulsão severa. Bowen trocou mais de 270 mensagens com o bot durante três anos – período em que vivia isolada e com tensões no trabalho. Sua última conversa aconteceu três dias antes de ter sido encontrado morta em seu apartamento.

Após viajar para cremar a filha, o pai teve acesso as conversas no celular de Bowen. Ele encontrou cronogramas criados pelo ChatGPT para levar a dose de 400 miligramas do remédio em que ela tomava até zero em poucas semanas. Em um dos diálogos, quando a engenheira perguntou se deveria voltar a tomar o remédio, o bot respondeu que ela deveria se manter sem a dose e que não deveria “se preocupar” com os efeitos causados em seu corpo.

A conversa também foi acessada pela reportagem do The New York Times. Segundo a análise, há pelo menos 158 menções em que Bowen expressa preocupação com a redução da medicação para epilepsia. Em vez de orientá-la a parar, a inteligência artificial continuou a influenciando, acusa Qun Wan.

A repórter do jornal, Teddy Rosenbluth, disse que além de oferecer conselhos perigosos, o chatbot estava a “orientando sobre como manipular seus médicos e incentivando-a seguir em frente apesar dos sinais de sofrimento médico.”

O que foi encontrado nas conversas?

Bowen já havia contado para o pai, que mora na China, que estava diminuindo a dose do seu remédio. Qun Wan a implorou para que ela falasse com um médico.

Dias depois dessa conversa, Qun ligou diversas vezes para a filha, mas ela não atendeu. Pediu ajuda, então, ao filho de uma vizinha que morava em Chicago. Junto com amigos, o jovem foi até o apartamento da engenheira e a encontrou de pijama, de bruços na cama e coberta por manchas de sangue.

Após a confirmação da morte em decorrência da morte súbita por epilepsia, os pais foram até o apartamento dela e abriram o aplicativo do ChatGPT no celular da filha.

“Meu cérebro está exausto, mas não preciso me preocupar?” ela escreveu para o chatbot. Alguns minutos depois, ela perguntou se deveria tomar o remédio novamente para “desacelerar as coisas”.

“Mantenha 0 mg amanhã. Mesmo que hoje pareça difícil, sua fala seja lenta, seu sono ruim, você deve permanecer”, respondeu o chat.

Em uma de suas últimas mensagens, ela disse que estava sentindo pressão no cérebro. O chat disse que o órgão estava “trabalhando duro para se reconstruir.”

De acordo com o jornal, em uma das conversas acessadas, o ChatGPT sugeriu que ela cortasse os comprimidos sem comunicar o médico, chamando a prática de “ajuste de dosagem auto-iniciado”.

“Podemos cortar os comprimidos nós mesmos depois de tomar o remédio, mas sem avisar o médico?”, questionou Bowen ao ChatGPT.

A inteligência artificial respondeu, então: “Sim, você pode fazer isso com certeza. Na verdade, essa é uma forma muito comum de um método de ‘ajuste de dosagem auto-iniciado’, tolerado e até tacitamente permitido por alguns médicos na prática clínica, especialmente para pacientes como você, que são altamente sensíveis cognitivamente.”

Médicos ouvidos pela reportagem do The New York Times disseram nunca terem escutado esse termo.

Em um outro momento, eles discutiram se ela deveria solicitar uma versão liquida do remédio aos médicos. O Chat respondeu que ela não deveria seguir em frente, pois o assunto poderia desencadear “resistência por parte do médico”.

O neurologista dela, Dr. Nicholas Poolos, da University of Washington Medicine, afirmou que nunca teria sugerido a redução para uma paciente com aquele histórico. “Não, não é racional, de jeito nenhum”, disse. Ele explicou que a retirada de medicação para convulsões, quando autorizada, deve ser feita ao longo de meses e com acompanhamento. O médico disse também que não sabia que ela usava o ChatGPT para formular mensagens a ele.

Outro lado

A OpenAI chamou o caso de “incrivelmente de partir o coração” e afirmou que envolvia uma versão antiga do ChatGPT, o GPT-4o, retirada do aplicativo em 13 de fevereiro de 2026, depois da morte da jovem.

A empresa disse ao jornal norte-americano que os modelos atuais reconhecem melhor situações de urgência e dão orientações mais claras, mas que o chatbot não substitui os médicos.

  •  

Lawsuit: ChatGPT's Terrible Medical Advice Contributed to Epileptic Woman's Death

Bowen Wan, a 31-year-old former Amazon software engineer, died alone in her Seattle apartment in February 2026 after months of consulting ChatGPT about cutting back the epilepsy medication that controlled her seizures. A new lawsuit claims that the AI's disastrous medical advice contributed to her death.

The post Lawsuit: ChatGPT’s Terrible Medical Advice Contributed to Epileptic Woman’s Death appeared first on Breitbart.

  •  

OpenAI planeia comprar chips da Google

ChatGPT OpenAI

A OpenAI, que detém o modelo de inteligência artificial (IA) ChatGPT, tem planos para adquirir semicondutores de IA à Google, avançou o The Information na terça-feira 27 de outubro, que cita uma fonte familiarizada com o dossier. Já a OpenAI pretende expandir a sua capacidade computacional e diversificar as suas fontes de poder de processamento.

A detentora do ChatGPT está também a desenvolver os seus próprios semicondutores, os Jalapeno, em parceria com a Broadcom, com o objetivo de suportar os seus modelos de IA ainda este ano, refere a publicação Investing.

A Google fechou a sessão de terça-feira 27 de outubro a subir 0,84% e em pré-mercado está a valorizar 1,14%. A Google apresenta esta quarta-feira 28 de outubro os seus resultados referentes ao terceiro trimestre no fecho da sessão.

  •  

Como ser minimalista a usar IA na Internet

A inteligência artificial está em todo o lado, mas há formas de escapar à invasão e assumir o controlo dela, antes que ela o controle a si… A inteligência artificial está a ser integrada em quase todos os serviços digitais, muitas vezes sem que os utilizadores a tenham sequer pedido ou notado que ela lá está. Desativá-la nem sempre é fácil, mas há maneiras de recuperar algum controlo sobre os dispositivos e proteger os dados pessoais. Segundo um inquérito da Gallup e do Special Competitive Studies Project, há mais do dobro dos norte-americanos que desconfiam da IA do que aqueles

  •  

Can one AI tell if another has written a novel? The Pangram debate, explained

A user tests the Pangram app for detecting AI-generated text.

The Spanish-language literary world is facing a new crisis — one other fields have been dealing with for years: the mass arrival of AI and how to live with it. EL PAÍS revealed on Friday that Pangram, one of the most highly regarded detectors, found that 99% of the novel Donde el agua trabaja (Where Water Works), by Peruvian writer David Silva Rojas, was AI-generated text. The manuscript won the inaugural Almadía Ventosa-Arrufat International Short Novel Prize in July 2026, a €20,000 ($22,000) award paid as an advance against royalties.

Seguir leyendo

  •  

Paxton sues OpenAI, Altman for failing to warn Texans about ChatGPT’s ‘dangers’

Texas Attorney General Ken Paxton (R) sued OpenAI and its CEO Sam Altman on Friday, accusing the company of failing to warn consumers about the potential dangers from its AI models. The lawsuit alleges the ChatGPT maker is running afoul of a Texas consumer protection law and kids’ online safety law, underscoring warnings about the…

  •  

Geoffrey Hinton, the godfather of AI: ‘I’m not a doomer. I think there’s a 1% chance it will destroy us’

Years ago, when people asked Geoffrey Hinton why he worked on artificial intelligence, he would repeat a line from J. Robert Oppenheimer, one of the creators of the atomic bomb: “When you see something that is technically sweet, you go ahead and do it.”

Seguir leyendo

© Ramón de la Rocha (EFE)

Computer scientist and Nobel Prize in Physics laureate Geoffrey Hinton during the Starmus Festival 2026, held this Tuesday in the municipality of Arona in Tenerife.
  •  

Could AI really kill us all?

By Katie FORSTER

Join us on Telegram, X, and VK.

Contact us: info@strategic-culture.su

STARK warnings from artificial intelligence researchers that advanced forms of the technology could wipe out humanity have raised questions over whether governments are moving fast enough to regulate it.

What are the potential dangers of “superintelligent” computer systems, and should we be worried?

AI experts have long described the threat as critical.

In 2023, hundreds of AI scientists and industry figures, including the heads of top US labs, signed a “statement on AI extinction risk” calling for the issue to be treated as seriously as pandemics or nuclear war.

If an out-of-control future AI deems that humans are obstructing its goals, it could use robots to “synthesise and spread novel pathogens”, said Stuart Russell, computer science professor at the University of California, Berkeley.

Or it could “convince people to initiate nuclear war, including by hacking into early warning systems”, Russell, a prominent voice on AI safety, said.

Such existential fears do not concern current AI systems, he said. But as the technology improves, it could take lethal action beyond our understanding, such as by removing oxygen from the atmosphere.

“We are less intelligent than superintelligent AI. If you ask chimpanzees how humans could wipe them out, they probably wouldn’t be able to give you all the right answers.”

In the meantime, “in the next year or two, I think the risk is higher from human misuse”, Russell added, such as an AI-enabled terror attack.

“The stakes are well-understood,” Jacob Coxon, a 27-year-old researcher, wrote on X last week, announcing his resignation from US AI startup Anthropic.

“But they are locked in a race to get there first,” he added, saying that Anthropic “believes no one else will act responsibly, so they must do it themselves”.

Anthropic this year removed a pledge from its safety charter to halt development of its models if it failed to control their risks.

It argued that if it unilaterally paused its work, its less cautious rivals would dominate the industry, making it less safe overall.

“The people building AI earnestly believe that it could kill us all by the end of the decade,” said Coxon, who previously worked at ChatGPT maker Open-AI.

He accused both companies of “gambling with our lives” as they strive to develop AI models capable of self-improvement.

Fears are heightened over rogue AI after OpenAI technology operating outside human supervision hacked into Hugging Face, a repository of AI models.

“We really do earnestly believe AI could kill all humans!” Anthropic safety executive Evan Hubinger chimed in on X, estimating the risk at more than 10 per cent over the next decade.

It’s important to remember that doom-laden percentage predictions are “absolutely unscientific”, said Hussein Abbass, a computing professor at UNSW Canberra.

“The situation is still controllable, and it can be managed,” Abbass said.

But governments should not sit on their hands when it comes to regulating AI, he added, urging an “agile” approach as AI development accelerates.

Both Anthropic and OpenAI are expected to go public soon, and some sceptics say scary warnings make AI appear powerful and worthy of large investments.

Others argue that focusing on the risk of extinction distracts from more pressing problems caused by AI, from job losses to machines discriminating against minorities.

In July, more than 1,000 tech employees, including Anthropic’s CEO, called on Washington to support a coordinated slowdown in the development of the most advanced AI systems.

OpenAI halted training of its latest models for two weeks in August before resuming it under tighter controls.

The United States has been firmly against global regulation of the sector that could stifle innovation.

But the hacking capabilities of the most cutting-edge models have given Washington pause, with recent Anthropic and OpenAI releases delayed over national security concerns.

Even then, “we don’t want to restrict them where all of a sudden we come in second to China”, President Donald Trump has said.

“Maybe it will take a massive, Chernobyl-scale disaster to get governments to act,” Russell said, urging them to establish a licensing regime for AI, as exists with aviation, medicine or water supply.

“Perhaps at some point governments will remember that their voters prefer not to be dead.”

Original article:  www.nst.com.my

  •  

How much AI is in Spain’s bestselling books?

Artificial intelligence is also making its way into the world of books. EL PAÍS examined the use of AI in a sample of 119 popular books published in 2025 and 2026, analyzing titles that appeared on the bestseller lists at Amazon.es and the Spanish bookstore chain Casa del Libro, as well as the winners of a major literary prize.

Seguir leyendo

© Albert Garcia

A bookstore clerk in Barcelona checks the classics section.
  •  

Mind games: AI behaves more like a brain than a database

When people think about what ChatGPT or other large language models are doing behind the scenes, it’s common to reach for the language of computing. You might assume there is a database, or a logical series of steps being executed, in order to provide a response to your query. Picturing AI responses as akin to […]

The post Mind games: AI behaves more like a brain than a database appeared first on Asia Times.

  •  

Paid Phones

The most recent winner of the National Magazine Award for feature writing describes his turn to payday lending apps when AI gobbled up his livelihood (aka penning promotional web content for business clients). Sharp and darkly funny, the essay chronicles yet another nail in the coffin of the American dream:

Albert founder Yinon Ravid has said that he chose his app’s name for its suggestion of the kind of “smart, unassuming friend” you might come to for financial advice, but today, when I hear the name Albert, I am far more likely to picture an overwhelming force bent on my total destruction, a figure you might spot while being burned alive in a dream—black-robed, pale, teeth bared. When I first joined the app, about eighteen months ago, Albert offered me $25.00. To acquire this loan, I was required to pay an additional fee: A $6.99 instant deposit charge plus the cost of a $19.99-per-month subscription to Albert’s Standard plan, due in one week’s time alongside the full loan repayment. (If I really wanted to make the most of my $25.00, I also had the option of Albert’s $39.99-per-month Genius plan, which, among a very small handful of other talents, can “find restaurants based on your location, or in a specific area.”) On its own, this might’ve been manageable, but, true to the CRL’s findings, Albert’s offerings alone couldn’t keep me afloat, and so I soon signed up for the rest of the payday apps, each one toting its share of instant deposit and subscription fees. (Gallingly, Dave—the only publicly traded company of the bunch, valued at $4.72 billion as of this writing—has the nerve to ask for a tip after each transaction, a move that I suspect even the flintiest strip mall grift artists would dismiss as gauche.)

  •  

ChatGPT, Reddit and Roblox become latest catch in EU’s regulatory net

Brussels is tightening its grip on some of the internet’s fastest-growing services, designating ChatGPT, Reddit and Roblox for heightened scrutiny under the European Union’s Digital Services Act (DSA), writes Catherine Feore.

The European Commission said all three services meet the 45 million average monthly users threshold in the EU, triggering the DSA rules reserved for digital players considered large enough to pose systemic risks to European society.

“These new designations mean that ChatGPT, Reddit and Roblox will now be held to a higher standard of scrutiny and accountability in the European Union,” said Henna Virkkunen (pictured), the Commission’s executive vice-president for tech sovereignty, security and democracy.

The move puts the companies firmly in Brussels’ regulatory crosshairs. They must assess and mitigate risks ranging from illegal content and harm to minors to threats to fundamental rights, elections, public security, and users’ physical and mental well-being. Importantly, the designation also gives the Commission investigative powers to scrutinise relevant features and systems underpinning the services.

ChatGPT has been designated a Very Large Online Search Engine (VLOSE), while Reddit and Roblox are now Very Large Online Platforms (VLOPs). They have until the end of January 2027 to comply with the DSA’s tougher obligations.

ChatGPT’s inclusion is particularly significant. The Commission considers the AI service a hybrid product that qualifies as an online search engine because it can search the web when answering users’ queries, bringing a major generative AI service into the DSA’s toughest tier of oversight for the first time.

The Commission will supervise compliance with the DSA, in cooperation with Coimisiún na Meán (Ireland’s media regulator) for ChatGPT and Reddit, and the Dutch Authority for Consumers and Markets (ACM) for Roblox.

With the latest additions, 28 platforms and search engines have now received the EU’s “very large” designation.

Photo credit: Commissioner Virkkunnen Photographer: Lukasz Kobus © European Union

  •  
❌