AI & Models
Anthropic shuts down AI models after U.S. government order
The U.S. government ordered Anthropic to immediately shut down two of its AI models worldwide, citing security concerns, a directive the company has complied with.
On Friday, the U.S. government ordered Anthropic to immediately shut off access to two of its AI models, Claude Fable 5 and Claude Mythos 5. Anthropic confirmed it complied with the government order. The shutdown directive forces the company to disable both models for all users worldwide, representing a significant escalation in regulatory scrutiny.
The government’s directive was issued as an export control order, but the underlying concern is a claimed jailbreak of Fable 5, which Anthropic released three days ago. A jailbreak refers to prompting an AI model to bypass its safety guardrails. According to Anthropic, the government has only provided evidence of a “potential narrow, non-universal jailbreak” for the model. The company disagreed that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. In a statement, Anthropic warned: “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.” The term frontier model refers to highly capable, state-of-the-art AI models. Fable 5 had recently been ranked as the most capable publicly available model by Vals AI, an AI performance benchmarking company.
The shutdown also affects Claude Mythos 5, a model Anthropic previewed in early April. Due to its capabilities, Anthropic had restricted Mythos 5 to Project Glasswing, which is a controlled access program. This program involves roughly 50 vetted organizations, which include Amazon, Apple, Google, Microsoft, and CrowdStrike. The regulatory action comes at a sensitive time for Anthropic, which is expected to pursue an IPO this year. The situation highlights the business risks of Anthropic’s safety-first positioning. Sam Altman, the CEO of rival AI company OpenAI, previously told podcaster Ashlee Vance that Anthropic’s strategy amounted to “fear-based marketing.” Altman argued that it is clearly incredible marketing to claim to have built a bomb and then offer to sell a bomb shelter for $100 million.
Why it matters
The U.S. government’s order to shut down Anthropic’s most powerful AI models represents a significant escalation in regulatory scrutiny over frontier AI safety. This action potentially sets a precedent that could halt future model deployments across the industry.