Anthropic AI Shutdown: Impacts and Controversies

Anthropic faces dilemma with shutdown of its most powerful AI

A Anthropic, famosa por suas preocupações com segurança, agora encara um revés. O governo dos EUA ordenou que a empresa desligasse imediatamente seus modelos de IA mais poderosos, o Claude Fable 5 e o Claude Mythos 5, alegando preocupações de segurança nacional. A Anthropic acatou a ordem, mas deixou claro seu desacordo. A diretiva, recebida numa sexta-feira à tarde, exige que a empresa desative os modelos para todos os usuários, não apenas para estrangeiros, como o controle de exportação sugeria. Outros modelos da Anthropic não foram afetados.

The impact of this decision is significant. Mythos is Anthropic's most advanced model, known for its ability to identify security vulnerabilities in software. Since its release, Anthropic has kept Mythos restricted, sharing it only with 50 selected organizations, such as Amazon and Google, for cybersecurity work. Fable 5, recently launched, is a version of Mythos with restrictions that block responses in high-risk areas, making it safe for general use. According to benchmark tests by Vals AI, Fable 5 was the most capable model available to the public.

The government's decision is presented as an export control action, but Anthropic believes the real concern is an alleged jailbreak of Fable 5. The company claims the government only provided verbal evidence of a "potential and non-universal jailbreak," which involves asking the model to read specific code and identify flaws. Anthropic argues that this capability is already available in other public models, such as OpenAI's GPT-5.5, and is routinely used by cybersecurity professionals.

Anthropic highlights that its strongest safeguards operate through independent systems, meaning that even if someone manages to keep Fable talking after a refusal, protections against dangerous outputs remain. However, this did not prevent the government action, and the company does not hide its frustration. "We disagree that the discovery of a potential jailbreak should be grounds to recall a commercial model deployed to millions of people," Anthropic wrote. "If this standard were applied industry-wide, we believe it would essentially halt all new model releases."

The irony is that Anthropic's caution in restricting Mythos, promoted as a model so dangerous that it could not be released publicly, attracted exactly the kind of government scrutiny that could harm its business. OpenAI's Sam Altman commented that Anthropic's approach looked like "fear-based marketing." He compared the situation to building a bomb and offering a shelter at an exorbitant price. Altman did not predict a government shutdown but pointed out that by spending months saying your AI is dangerously unique, the world, including the US government, tends to listen.