US Government Export Control Directive Suspends Fable 5 and Mythos 5 for Foreign Nationals

On June 12, 2026, the US government issued an export control directive requiring Anthropic to suspend access to its two newest models, Fable 5 and Mythos 5, for all foreign nationals — including foreign national Anthropic employees — citing a national security concern about a demonstrated jailbreak [2]. Anthropic is complying under legal obligation while publicly contesting the technical basis, arguing the jailbreak amounts to asking the model to read a codebase and fix flaws, a capability it says is widely available from other models including OpenAI's GPT-5.5 [2][3]. The government has not disclosed the specific nature of its national security concern [2].

This appears to be the first instance of export control authority being applied to pull a commercial AI model after launch. Anthropic argues that if a narrow potential jailbreak is sufficient grounds for recall, the same standard applied industry-wide would halt all new frontier model deployments — a claim that, if accurate, would make this directive a significant constraint on how AI models can be commercially deployed in the US [2].

On June 12, 2026, the US government — identified by Axios as the Trump administration via Commerce Department authority — issued an export control directive requiring Anthropic to suspend all foreign national access to Fable 5 and Mythos 5, its two most recently released models [1][2]. The directive cited a jailbreak as the national security basis but did not publicly specify the nature of the concern [2].

Anthropic published a statement the same day acknowledging it is complying under legal obligation while contesting the technical merits of the action. The company's review of the demonstrated vulnerability found it consists of asking the model to read a codebase and identify flaws — a capability it describes as widely available from other models including OpenAI's GPT-5.5, with no Mythos-specific uplift [2][3]. Anthropic noted it had conducted thousands of hours of pre-launch red-teaming with government and private organizations yielding no universal jailbreak, and had deliberately built 30-day data retention into Fable deployments specifically to enable rapid detection and mitigation of successful attacks [2].

On process, Anthropic argued the action is disproportionate and lacks transparency, and called for a statutory AI deployment review process that is fair, clear, and grounded in technical facts [2]. The company's sharpest public claim is that if the standard — blocking a model over any narrow potential jailbreak — were applied consistently across the industry, it would effectively halt all new frontier model deployments by every provider [2].

Simon Willison, reporting on June 13, noted that he personally still had access to Fable via Claude.ai and Claude Code at 9:01pm ET, despite the directive having been issued — pointing to either an enforcement lag or incomplete access control rollout [3]. Coverage spread rapidly across technology media, Reddit, and Twitter, though substantive reaction from other AI companies or government officials has not yet surfaced in available items [4][5][6].