Anthropic has pulled its flagship AI models, Fable 5 and Mythos 5, offline across all global markets after a directive from the White House raised alarms over national security vulnerabilities. The order demanded immediate suspension of access for all foreign nationals, regardless of location, including Anthropic’s own employees abroad. This unprecedented intervention underscores escalating government scrutiny over AI technologies amid geopolitical tensions.
The White House’s concerns centered on fears that the models could be exploited through unknown access pathways possibly linked to China-affiliated entities. While Anthropic maintains it engaged extensively with U.S. and U.K. security agencies in testing the models, the government reportedly identified a method to bypass safety guardrails in Fable 5. Anthropic investigated this lead and concluded that any discovered vulnerabilities were limited and non-critical, but the broader risk prompted the decisive government action.
The conflict between Anthropic and U.S. federal agencies has been intensifying for months. Earlier in the year, federal authorities halted all use of Anthropic technology, and the Pentagon labeled the company as a supply-chain risk, largely due to disagreements over the use of AI for autonomous weaponry and mass surveillance. Anthropic resisted these applications on ethical grounds, arguing its models were unsuited for lethal autonomy and worried about infringing on civil liberties.
Despite tentative government efforts to ease restrictions in the spring, the launch of Fable 5 and Mythos 5 reignited tension. In response to the shutdown order issued on June 12, Anthropic deactivated the models for every user worldwide but kept other AI products online. This move demonstrates growing concerns about how emerging AI systems intersect with international security and export control regimes.
This episode signals a new chapter in the regulatory landscape for advanced AI, where national security considerations may override commercial and innovation interests. It also highlights the complexities faced by AI developers navigating both technical safety and geopolitical risk in a rapidly evolving environment.