Anthropic announced on Friday that it would "abruptly disable" its most advanced AI models for all users following a US government order to suspend access for foreign nationals due to national security concerns. The company received an export control directive to block access to Fable 5 and Mythos 5 for all foreign nationals, without specific details about the security threat, according to an Anthropic statement.
Government Concerns Over Jailbreaks
Anthropic understands that the government believes there is a method to bypass safeguards, known as "jailbreaking," which could allow Fable 5 to be used for identifying software vulnerabilities. The order comes as tensions between the Trump administration and IPO-bound Anthropic were showing signs of easing. The relationship deteriorated earlier this year when Anthropic refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems. In response, the government placed Anthropic on a supply chain blacklist, effective later this year.
Escalation of US Export Controls
This action marks a significant escalation in US efforts to curb foreign adversaries' AI capabilities. Previously, US export controls targeted chips and tools powering AI, not restricting foreign access to AI itself. Anthropic stated it received only "verbal evidence of a potential narrow, non-universal jailbreak" and disagreed that this should warrant recalling a commercial model used by hundreds of millions.
The directive and Anthropic's response highlight growing friction between AI developers and regulators over assessing risks from jailbreaks. As recently as Wednesday, Anthropic called for greater US oversight of AI, including blocking models with unacceptable risks, but claimed Friday's action violated principles of fair and fact-based regulation.
Pentagon and Industry Reactions
Pentagon CIO Kirsten Davies posted on X that the Defense Department supports prioritizing national security over "revenue cycles, clickbait, and pre-IPO valuation." Anthropic confidentially filed for a US IPO last month, ahead of rival OpenAI. Earlier this week, Anthropic launched Claude Fable 5, a "Mythos-class" model with guardrails barring risky uses like cybersecurity, which some users deemed overly broad.
Experts warn that Mythos models could accelerate sophisticated cyberattacks in sectors like banking. Anthropic noted it worked with the US government on safety before the launch and that rival models show similar abilities to find code bugs. The company said the order forces it to disable Fable 5 and Mythos 5 for all customers, while other models remain unaffected.
Misunderstanding and Industry Impact
Anthropic believes there is a "misunderstanding" and is working to restore access. It warned that if this standard were applied industry-wide, it would halt all new model deployments for frontier providers. Amazon's AWS confirmed it revoked access to the models for all users globally at Anthropic's request. A US official confirmed the Commerce Department issued the export control directive.
Dean Ball, a former White House official, posted on X that the order suggests all non-Americans, including those in the US, would be restricted from using Anthropic's latest models, potentially requiring proof of citizenship. Key Anthropic personnel, including co-founder Chris Olah, researcher Andrej Karpathy, and philosopher Amanda Askell, were born outside the US. Their citizenship status remains unclear, and Anthropic declined to comment on whether they would lose access.
