Four days after launch, Anthropic yanked both Fable 5 and Mythos 5 offline for every user. The White House ordered it on June 12, and Anthropic complied, but not without a pointed farewell: "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."
The Models That Couldn't Stay Live
Anthropic launched Fable 5 on June 9, calling it "the most capable model we've ever made generally available." Mythos 5, released simultaneously, shared the same underlying weights but with safety guardrails partially lifted. That distinction mattered the moment security researchers found a jailbreak that could make Fable 5 serve information useful for cyberattacks.
Amazon's Security Research and the White House's Order
The order didn't come from thin air. According to reports, Amazon relayed the research findings to the White House after its own security team confirmed the vulnerability. Anthropic said the government's legal directive required cutting access to both models, even though the reported exploit targeted only one variant.
Anthropic's Compliance and Quiet Rebuke
Anthropic's response is notable for what it admits and what it challenges. The company shut off both models immediately, acknowledging the legal obligation. But the public statement frames the recall as disproportionate, arguing that a single jailbreak path doesn't justify pulling a widely deployed commercial system. The subtext is clear: if this is the bar for a federal model recall, every frontier lab with a bug bounty program could face similar shutdowns.
This episode sets a precedent for how the U.S. government might handle future jailbreak discoveries, especially when cloud providers like Amazon are in the loop. Expect every AI company with a government contract to revisit its launch-and-patch playbook.
Source: All the news about Anthropic's new AI fight with the White House
Domain: theverge.com
Comments load interactively on the live page.