US blocks access to Anthropic AI models, halting global development

Developers worldwide faced disappointment as Anthropic abruptly took its latest AI models, Fable 5 and Mythos 5, offline just days after their release. The US government's decision to prohibit non-Americans from accessing Fable 5 prompted Anthropic to disable both models globally, effectively halting their development and use.

Mythos 5, designed for identifying cybersecurity weaknesses, is restricted to a select group of governments and companies. Fable 5, a version of Mythos 5 with added safety features, was intended for broader availability. However, the US government cited evidence that both models could be easily "jailbroken" – meaning their internal rules could be bypassed to perform unintended functions. This vulnerability, according to the US, could allow Fable 5 to behave like the more restricted Mythos 5, potentially posing security risks. Anthropic has reportedly denied the severity of these jailbreaking issues.

The block represents an escalation in the global race for advanced AI technology. Anthropic had already implemented access restrictions, initially limiting Mythos 5 to US institutions before expanding to a limited international group. This move follows a recent US executive order by President Donald Trump mandating government review of AI models that could function as cyber weapons before their launch. The broader implications remain unclear, with questions arising whether this ban will extend to other major US AI companies or if it is specific to Anthropic due to its strained relationship with the US government.

The situation is expected to intensify pressure on Europe to accelerate its own AI model development. For companies already working with these models, the immediate impact might be manageable. Piet Kerkhofs, CTO of Eye Security, a European cybersecurity firm, noted that his company, which has used AI for four years, can easily switch to alternative models. He stated, "What used to take us a month..." indicating that the rapid advancement of AI allows for quick adaptation, even with such disruptions.

“

jailbreaken