Anthropic says it was compelled to shut down entry to its new Claude Fable 5 and Claude Mythos 5 AI models after the Trump administration issued a broad warning towards use of the fashions by overseas nationals. The warning, from the Commerce Division, would have banned even a lot of Anthropic’s personal staff from utilizing or engaged on the fashions.
That dramatic end result got here after a multiday back-and-forth involving quite a few administration officers and enter from tech trade leaders. The consequences of the federal government’s successfully shutting off entry to 1 firm’s AI fashions might have profound implications for future AI coverage. What follows is a timeline of the occasions that led to the efficient ban on Anthropic’s Mythos-class fashions.
Thursday evening (June 11)
Thursday night: Amazon raises considerations with Trump administration officers
Amazon, Anthropic’s greatest investor, contacted senior White Home and administration officers, saying its researchers had recognized what they believed was a strategy to coerce, or “jailbreak,” the Fable 5 and Mythos 5 fashions into helping customers with dangerous cybersecurity questions. Anthropic launched the fashions with safeguards designed to forestall responses to such queries. Axios reported that a minimum of 5 different tech corporations raised comparable considerations with administration officers across the identical time.
Late Thursday evening: Administration begins treating the difficulty as pressing
White Home and Commerce Division officers took Amazon’s considerations significantly, worrying that overseas unhealthy actors would possibly bypass Anthropic’s safety guardrails and use the fashions to determine and exploit software program vulnerabilities, Axios reported. The White Home started attempting to contact Anthropic, asking for a direct reply.
Friday morning (June 12)
Friday morning: White Home and Commerce officers interact Anthropic
Administration officers reportedly spent hours Friday attempting to persuade Anthropic to voluntarily shut off entry to the Fable 5 and Mythos 5 fashions. It’s unclear from the reporting whether or not the federal government additionally requested Anthropic to repair the particular security guardrail that allowed Amazon researchers to jailbreak the system, or to voluntarily limit overseas entry to the fashions.
Friday morning: Anthropic declines to voluntarily droop the fashions
Axios reports that the federal government tried to steer Anthropic to pause the discharge of the Fable 5 and Mythos 5 fashions, however was unsuccessful. Anthropic has mentioned publicly that the federal government didn’t present detailed technical proof of the alleged subject and that the considerations concerned solely verbal descriptions.
“We reviewed an indication of this particular method getting used to determine a small variety of beforehand identified, minor vulnerabilities,” Anthropic mentioned in a Friday, June 12, blog post. “These vulnerabilities all seem comparatively easy, and we’ve discovered that different publicly out there fashions are in a position to uncover them as effectively with out requiring a bypass.”
Friday afternoon
Round 1 p.m. ET: ‘You’ve gotten 90 minutes to conform’
Anthropic obtained a name from administration officers round 1 p.m. ET, warning that until the corporate acted, the federal government would transfer ahead with export restrictions affecting the fashions, Reuters and Axios reported. Axios additional reported that Anthropic was given roughly 90 minutes to conform.
5:21 p.m. ET: Commerce Division order arrives
At 5:21 p.m. ET Friday, the Commerce Division despatched an enforcement order limiting entry to Fable 5 and Mythos 5 by overseas nationals, citing nationwide safety considerations.
Friday night and evening
Friday night: Anthropic disables Fable 5 and Mythos 5
Anthropic mentioned it disabled entry globally as a result of the order restricted entry by overseas nationals and the corporate couldn’t instantly separate eligible from ineligible customers. It erred on the facet of warning and compliance.
Context
About Fable 5 and Mythos 5
Fable 5 and Mythos 5 are the primary publicly launched fashions derived from Anthropic’s Mythos household. The unique Mythos mannequin confirmed uncommon talent throughout coaching at discovering software program bugs and exploiting them to disrupt or take management of programs. That raised sufficient concern inside Anthropic that the corporate grouped cybersecurity with different high-risk domains, together with biology and chemistry, when setting limits on Mythos-derived public fashions. For Fable 5 and Mythos 5, meaning prompts flagged as delicate in these areas are routed to Claude Opus 4.8, a much less succesful mannequin with its personal guardrails.
In regards to the politics
Many observers noticed the tech trade’s shocking assist for Donald Trump’s presidential candidacy in 2024 as a part of an implicit discount over AI coverage. Trump pledged that his administration would permit the AI trade to control itself and do what it might to advance the tech trade’s aim of stopping states from passing their very own AI legal guidelines within the absence of federal laws. The Trump administration has additionally ordered federal businesses to drop numerous investigations and enforcement actions involving huge tech corporations. So the administration’s current considerations in regards to the nationwide safety dangers of the most important AI fashions, and its pivot towards involvement in safeguarding fashions like Mythos, is putting. (See: Trump’s June 2026 executive order.)
The present dustup may also be partisan and political. In March, Protection Secretary Pete Hegseth clashed with Anthropic after the AI firm refused to permit its AI for use for autonomous weapon concentrating on or home surveillance. The Pentagon then designated Anthropic a “supply-chain danger,” successfully blacklisting it from a lot federal protection work. Anthropic sued.
In regards to the fashions
Earlier than launch, Anthropic mentioned inside and exterior purple groups spent greater than 1,000 hours attempting to idiot Fable 5 into shelling out info on banned subject areas. Anthropic mentioned the groups had no success figuring out common jailbreaks.
Nonetheless, a minimum of one impartial AI researcher seems to have discovered a successful jailbreak earlier than Amazon researchers went to the federal government with their report. The researcher, utilizing the deal with Pliny the Liberator, claimed on X to bypass Fable 5’s filters utilizing a multi-agent method involving a beforehand jailbroken Claude Opus 4.8 mannequin, together with strategies together with question decomposition, long-context framing, fiction and narrative constructions, and educational taxonomies.
