Trump administration blocks foreign access to Anthropic AI models under export controls

Trump administration blocks foreign access to Anthropic AI models under export controls

Anthropic kept lower-capability models online after disabling Fable 5 and Mythos 5, a move that drew industry criticism and may weigh on sentiment around its confidential IPO filing.

TRUMP

Fact Check
Anthropic's own official statement and X post confirm a US government export control directive (attributed to the Commerce Department by WIRED and Fortune) that required licenses/suspension for export, reexport or transfer to foreign nationals — including the company's own foreign-born employees — prompting Anthropic to deactivate Claude Fable 5 and Mythos 5 for all customers. Independent primary reporting from WIRED and Fortune (and additionally WSJ and Washington Post) corroborate. The framing as occurring 'under the Trump administration' aligns with reporting describing escalating tensions between Anthropic and the Trump administration.
Summary

Anthropic disabled Claude Fable 5 and Claude Mythos 5 on June 12 after a U.S. Commerce Department order required licenses for any export, reexport or transfer of the models to foreign individuals, including inside the United States and among Anthropic’s own foreign-born employees. Anthropic said it could not effectively enforce nationality-based restrictions and therefore shut the models down entirely, while lower-capability models such as Claude Opus 4.8 remained available. The action followed administration concerns, reported by Axios and The Wall Street Journal, that the models posed security risks tied to a narrow technique that could bypass part of Fable 5’s safety system, though Anthropic said the issue did not amount to a universal jailbreak and exposed only previously known minor vulnerabilities. Fortune reported on June 13 that the shutdown triggered criticism across the AI industry and could weigh on investor sentiment as Anthropic pursues a confidential IPO filing after its latest funding valued the company at $96.5 billion.

Terms & Concepts
  • export controls: Government restrictions that limit how sensitive technology can be shared, transferred or accessed across borders or by certain individuals.
  • universal jailbreak: A broadly effective method for defeating an AI model’s safety controls across many different tasks or capabilities.
  • red-teaming: Structured stress-testing in which researchers probe a system for weaknesses, misuse risks or safety failures before or after release.