Claude Mythos 5 redeployed to selected US organisations

Suspended since 12 June, Anthropic’s Claude Mythos 5 model has been redeployed. However, it is currently available only to a small group of organisations, all of which are based in the United States. Talks between Anthropic and the US government nevertheless appear to be progressing constructively: the AI lab is now discussing a broader rollout of Claude Mythos 5 to additional organisations, alongside the return of Claude Fable 5 for the general public.

Anthropic had paused Claude Mythos 5 on 12 June in order to comply with a US government directive. The powerful AI has now returned for selected organisations, according to an announcement from its developer published on X. That said, the renewed deployment remains restricted to a limited set of organisations-and, crucially, only US organisations are included.

Claude Mythos 5 returns for selected US organisations

“Today, the government informed us that Mythos 5, our highest-performing cybersecurity model, could be redeployed to a set of US organisations responsible for operating and protecting critical infrastructure,” the announcement states. Rumours also suggest that Claude Fable 5-the public version of Claude Mythos 5-could return this week.

As a reminder, Claude Mythos 5 is a highly capable cybersecurity AI that can identify hundreds of security flaws in software. It is reserved for selected organisations, which use it for defensive purposes.

Fable 5 could return as well

Claude’s developer also says it is continuing to work with the US government to widen access to Claude Mythos 5 and make Fable 5 available to the public again. Claude Fable 5 is the consumer version of Claude Mythos 5. Although Fable 5 offers the same performance levels as Mythos 5, its capabilities in cybersecurity, biology and chemistry have been restricted for safety reasons.

Why Claude Fable 5 was suspended

Fable 5 was suspended after the US government became convinced that a method already existed to remove Claude Fable 5’s restrictions and use it for malicious purposes. Anthropic, however, maintains that this technique does not genuinely bypass Claude Fable 5’s safeguards.

“We reviewed a demonstration of this specific technique being used to identify a small number of already-known minor vulnerabilities. These vulnerabilities all appear to be relatively straightforward, and we found that other publicly accessible models can also identify them without requiring a workaround,” Anthropic argued in a blog post.