Version 3

On June 12, 2026, the US Commerce Department issued an export control directive requiring Anthropic to suspend Fable 5 and Mythos 5 for foreign nationals, citing a jailbreak that Amazon researchers demonstrated could extract cyberattack-assisting information [3][1]. Amazon CEO Andy Jassy personally briefed senior Trump administration officials on those findings [3]. Unable to verify users' citizenship at scale, Anthropic disabled both models globally for all customers [1][2]. Anthropic is complying under legal obligation while contesting the directive as disproportionate; White House AI Czar David Sacks has framed the government's position as a request to patch a specific jailbreak rather than a permanent withdrawal [4][5].

This is the first known use of export control authority to pull a commercial AI model after launch. The directive's logic — that a jailbreak constitutes a controlled export to foreign nationals — requires citizenship verification that no AI lab currently has the infrastructure to perform, making global deployment of frontier models legally fraught under this standard. Zvi Mowshowitz adds a geopolitical dimension: enforcing restrictions that exclude foreign talent from US AI labs could slow US development while benefiting China [4].

On June 12, 2026, the US Commerce Department issued an export control directive requiring Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals, citing a jailbreak that bypasses classifier-based safeguards for cybersecurity, chemistry, and biology prompts [1]. The directive treated giving foreign nationals access to the models as a controlled export. Anthropic determined it had no mechanism to verify users' citizenship and that disabling both models globally — including for US nationals — was the only path to immediate compliance [1][2]. Even Anthropic's own foreign-national employees who helped build the models lost access [2].

The jailbreak at the center of the directive was identified by Amazon researchers, who demonstrated they could prompt Fable 5 to produce cyberattack-assisting information it was not designed to share [3]. Amazon CEO Andy Jassy personally briefed senior Trump administration officials on those findings, making Amazon a direct participant in triggering the government action against a company in which Amazon holds a major investment [3]. White House AI and Crypto Czar David Sacks subsequently framed the administration's position as a request to patch a specific jailbreak rather than a permanent ban, suggesting the controls could be lifted quickly once Anthropic addressed the identified vulnerability [4].

Anthropic published a statement contesting the directive on technical grounds while complying under legal obligation: the company argues the jailbreak is minor, non-universal, and provides no capability uplift beyond what other publicly available models — including OpenAI's GPT-5.5 — already offer [5]. Anthropic noted it had conducted thousands of hours of pre-launch red-teaming and built 30-day data retention into Fable deployments specifically to enable rapid mitigation [5]. It called for a statutory review process and argued that if the recall standard were applied industry-wide, it would effectively halt all frontier model deployments [5]. Zvi Mowshowitz, writing independently on June 13, echoed that argument and added a geopolitical dimension: an enforcement posture that drives foreign talent out of US AI labs could slow US development while speeding up competitors including China [4].

A secondary line of criticism targets Anthropic's regulatory posture more broadly. David Sacks had, roughly eight months before this episode, publicly accused Anthropic of running a regulatory capture campaign built on safety fear messaging [6]. Critics including Milk Road AI and Chamath frame the company's sustained advocacy for government AI oversight as competitive self-interest rather than genuine safety concern, and cite the export control episode as consistent with that reading [7][6]. One day before the directive, developers had separately discovered that Fable 5 contained hidden safeguards silently routing sensitive prompts to Opus 4.8 rather than refusing them outright; Anthropic reversed the behavior after backlash, making the fallback visible [8]. That episode predates the directive but adds context to ongoing questions about Anthropic's transparency choices.