On June 13, the Trump administration imposed export control directives on two flagship AI models from Anthropic — Fable 5 and Mythos 5 — barring any foreign government, business, or individual (including Anthropic’s non-U.S. employees) from accessing the models. To ensure compliance, Anthropic immediately took both models fully offline. Subsequently, co-founder and chief computing officer Tom Brown and public policy lead Sarah Heck were stationed in Washington, holding daily online meetings with government officials including Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross. The immediate trigger for the ban was reportedly intelligence suggesting a China-linked group may have accessed Mythos 5, as well as the discovery by Amazon researchers of a jailbreak path for the model.
On the government side, White House AI policy director David Sacks publicly accused Anthropic CEO Dario Amodei of prioritizing commercial deployment over safety and lacking cooperation — Amodei had engaged in three rounds of calls with multiple officials, arguing that the security vulnerability in question did not constitute a true “jailbreak” that would allow bypassing guardrails. Anthropic subsequently issued a statement saying it had only received oral evidence of a “potential, narrow and non-general jailbreak method,” and that the vulnerability was “relatively simple,” with similar weaknesses discoverable in other public models without the need for a jailbreak. The statement warned that “if this standard were universally applied across the industry, it would effectively halt all frontier model providers from deploying new models.” Reports indicate both sides are willing to reach a compromise, with direct engagement between Anthropic engineers and government security researchers seen as the most likely path to resolution. This also marks the first time a leading U.S. AI company has been forced to negotiate with the government under export controls to secure market access for its most advanced models.