Anthropic said on Friday it would “suddenly disable” its most advanced forms of Artificial Intelligence (AI), The legend of Claude 5 again Myths 5to all users after the US government ordered it to stop access to the models by foreigners, whether inside or outside the US, citing national security concerns.
The AI company said it received an order at 5:21 pm ET, ordering it to stop all access to the models by outsiders. It said it believed there was a “misunderstanding” and was working to restore access to the models as soon as possible. Access to other models will not be affected by the export control order.
“Our understanding is that the government believes it already knows how to bypass, or ‘jailbreak’ Fable 5,” the company said.
“We reviewed the demonstration of this particular technique used to identify a small number of previously known small vulnerabilities. These vulnerabilities appear to be relatively simple, and we found that other publicly available models are able to detect them as well without requiring a bypass.”
This unexpected move comes days after the launch of the Claude Fable 5 and its counterpart Mythos 5, which use the same basic model but with increased protections in other areas, such as cyber security. The latter, described as “the strongest cyber security force in the world,” remains available to a vetted group of cyber defenders and critical infrastructure operators.
Anthropic stressed that it has implemented “rigorous” safeguards to prevent misuse of its models in cyber security-related activities. In particular, this is bound under a set of security separators that are used to detect possible abuses, including jailbreak attempts, and prohibiting the main model from responding.
The cybersecurity category is designed to prevent one-off dangerous applications related to planning cyber attacks, exploit development, or evading defenses, the company notes that Mythos category models have the ability to find and exploit software vulnerabilities, thus giving attackers a strategic advantage.
Last week, Anthropic revealed that its Mythos-class model can turn newly disclosed software vulnerabilities into hours, or even minutes in some cases, instead of weeks, turning N-days into N hours. The findings suggest that boundary models may be good at quickly exploiting publicly disclosed errors.
“A freelancer can now turn a month’s worth of work into one afternoon’s worth of work – for a few thousand dollars without special expertise,” says Anthropic’s Red Team. “This means that the patching playbook used by software developers today – with monthly release cadences, multi-week staged releases, and lag between pre-release and stable channels – no longer holds.”
The protection of Fable 5 means that questions on cybersecurity topics will instead receive an answer from Claude Opus 4.8, the company’s next active model.
In its latest statement, the company pointed out that no jailbreak methods have been developed against the latest models so far, adding that third-party and internal tests by the red team found their defenses “much more effective than any model used before.”
In addition, Anthropic said that “complete resistance to jailbreak” is impossible for any provider model, since all the protections used by the industry may be vulnerable to non-standard jailbreaks that “work well in very limited situations or require more effort to adapt to each new situation.”
“So far, the government has only given us verbal evidence of a potentially small, non-universal jailbreak, which involves asking the model to read a specific codebase and fix any software errors,” he said.
“Our understanding is that this is one jailbreak that may have been shared with the government. We reviewed the report that we believe is the basis of the government order and confirmed that the level of capability demonstrated there is widely available in other models (including OpenAI’s GPT-5.5), and is used daily by defenders who keep systems safe.”
Anthropic also pointed out that although the government is all about preventing the deployment of unsafe AI, it said that the discovery of a “small jailbreak” should not be a reason to recall the widely distributed commercial model. The legal process should be “transparent, fair, clear, and based on technical facts,” it added.
Earlier this year, the US Department of Defense labeled Anthropic a “supply chain risk” after Claude’s maker sought to draw red lines for military use of its technology. The company filed two lawsuits to block the appointment.
