A Anthropic suspendeu o acesso a seus novos modelos de inteligência artificial, Fable 5 e Mythos 5, para usuários estrangeiros. A medida teria sido tomada por ordem de determinação do governo dos Estados Unidos. A decisão vale para qualquer cidadão de outro país, mesmo que esteja dentro do território americano, e também afeta funcionários estrangeiros da própria empresa.
Segundo a companhia, a ordem foi emitida com base em questões de segurança nacional e não especifica quais riscos teriam motivado a decisão.
Em comunicado, a Anthropic disse entender que as autoridades acreditam ter identificado uma forma de contornar as proteções de segurança do Fable 5, mas afirmou que a técnica apresentada envolve vulnerabilidades consideradas limitadas e que capacidades semelhantes podem ser encontradas em outros modelos disponíveis publicamente.
A empresa também declarou que analisou uma demonstração da suposta técnica e encontrou apenas falhas menores, sem benefícios específicos relacionados ao modelo. Nenhum teste teria encontrado até agora um método universal capaz de quebrar as barreiras de segurança do sistema.
A Anthropic defendeu que o Fable 5 foi desenvolvido com mecanismos avançados de proteção para reduzir o uso indevido da tecnologia em atividades relacionadas a ataques digitais. Segundo a empresa, o modelo passou por milhares de horas de testes de segurança antes do lançamento, incluindo avaliações com órgãos governamentais e organizações privadas.
Os demais modelos do Claude continuam funcionando normalmente. O governo dos EUA não fez maiores declarações.
Fundada em 2021 por ex-funcionários da OpenAI, a Anthropic é uma das principais empresas da corrida pela inteligência artificial generativa. A companhia é responsável pelo chatbot Claude, que oferece recursos semelhantes ao ChatGPT.
Apesar de ser uma empresa recente, a Anthropic recebeu investimentos bilionários de gigantes da tecnologia, como Google e Amazon, e ganhou destaque por defender uma abordagem voltada à segurança no desenvolvimento de sistemas de IA.
A empresa afirma que sua missão é criar modelos mais confiáveis e com maior controle sobre possíveis usos indevidos.
Jovens estão entre os mais desconfiados com IA, diz pesquisa global
Anthropic, one of the world’s leading artificial intelligence companies, has called for a slowdown in the development of advanced AI systems, warning that humanity may be approaching a point where the technology becomes difficult to control.
The company behind the Claude chatbot said it would support a temporary pause in developing more powerful AI models if other leading developers agreed to do the same.
Anthropic argued that a slowdown could provide governments, researchers, and society with more time to understand and manage the risks associated with increasingly capable AI systems.
The warning comes as Anthropic continues to position itself as one of the industry’s strongest advocates for AI safety. The company has reportedly withheld public access to its most advanced AI system, known as Mythos, because of concerns that it could be misused for large-scale cyberattacks and other harmful activities.
In an essay published Thursday, Marina Favaro, head of Anthropic’s research division, and company president Jack Clark said AI may be approaching a critical milestone known as “recursive self-improvement.”
The concept refers to AI systems helping design and improve future generations of AI with decreasing human involvement. Researchers said such a development could rapidly accelerate technological progress but also introduce new challenges for oversight and control.
LATEST: Claude maker Anthropic is calling for a global pause in AI development, warning that models are approaching the ability to self-improve without human intervention. pic.twitter.com/7WM9jmDZjt
— Cointelegraph (@Cointelegraph) June 4, 2026
Favaro and Clark pointed to growing evidence that AI is already contributing to its own development. According to the company, employees now produce roughly eight times more code than they did between 2021 and 2025, largely because of AI-assisted software development. The researchers also said AI systems are becoming better at generating ideas, planning research, and supporting scientific work.
Anthropic Chief Executive Officer Dario Amodei has previously warned about the potential risks of advanced AI, estimating there is a 25% chance that the technology could lead to severe negative outcomes if it is not developed responsibly.
Despite advocating caution, Anthropic acknowledged that coordinating a worldwide slowdown would be extremely challenging.
The company said any meaningful pause would require cooperation among leading AI laboratories across multiple countries. It also noted that verifying compliance would be difficult because AI training runs can be conducted privately and are far less visible than traditional military infrastructure.
Anthropic compared the competition to develop advanced AI to an arms race, arguing that companies may feel pressure to move faster to avoid falling behind rivals.
Not all experts agree with Anthropic’s assessment. Some researchers and industry observers argue that AI companies may be overstating the capabilities of current systems or emphasizing risks to encourage regulations that could disadvantage competitors.
The debate comes amid growing uncertainty in the AI sector. Shares of semiconductor company Broadcom recently fell sharply after disappointing investors with its sales outlook, triggering a broader decline among AI-related chipmakers and renewing concerns that enthusiasm surrounding artificial intelligence may have outpaced market realities.
As AI capabilities continue to advance, questions about safety, governance, and oversight are expected to remain central to the global conversation surrounding the technology.
Login