Anthropic, the company behind the Claude AI assistant, launched two new models on June 9: Claude Fable 5 for the public, and Claude Mythos 5, a more powerful version kept restricted because of how good it is at finding security holes in software. Three days later, Anthropic cut off access to both, after the US government issued an export control order limiting their use to American citizens.
The trigger was a discovered jailbreak (a trick that gets an AI to ignore its own safety rules) that could pull out information useful for real cyberattacks, and officials reportedly worried it could help hackers working for other countries. Picture a lock-picking kit so effective that even the people who built it got nervous about who might buy one. Talks between Anthropic and the government broke down, and the Department of Defense reportedly labeled the company a "supply chain risk," a label usually reserved for foreign adversaries.
On June 26, the restriction eased. Commerce Secretary Howard Lutnick said Mythos 5, still not Fable 5, could go out to more than 100 vetted US companies and agencies that protect critical infrastructure. Anthropic says it's still pushing for full availability. If you use Claude day to day, nothing changes for you: this dispute is about Anthropic's newest, most restricted tier, not the version most people actually use.
A more capable, more dangerous version of Claude collided with US national security policy, and the fallout is still being sorted out.