Anthropic Suspends Flagship AI Model After US Security Order
Anthropic has halted the public operation of its new AI models, Claude Fable 5 and Mythos 5, after a U.S. government directive cited potential security vulnerabilities. The order came just days after the tools went live, prompting the company to stop all access for foreign nationals to ensure compliance.
In a statement on its website, Anthropic explained that the government had directed the suspension of the two models because it considered them “too powerful.” The company has no public record of specific security concerns, but officials think the AI could be bypassed—often called a jailbreak—to read sensitive data or activate hidden features.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
The announcement comes amid a lawsuit that Anthropic and the Trump administration are already facing in the United States over a Pentagon order that labels the company’s tools a supply‑chain risk. The defense secretary’s remarks marked the first time a U.S. agency publicly offered that designation.
Claude Fable 5, a variant of the company’s Claude Mythos line, competes with OpenAI’s ChatGPT and Google’s Gemini. It was initially released privately to a small cohort of firms in April, where developers tested for potential vulnerabilities. Anthropic said it did not find “simple” or “minor” exploits—though the new models still rely on rigorous safeguards that many critics believe to be inflated marketing.
The company maintains that its capabilities now exceed any model it has previously released. With the U.S. court refusing to enforce the Pentagon’s directive, federal agencies still use Anthropic’s tools while litigation moves forward.
Key takeaways: the order stalls public use, warns against potential jailbreaks, and sets a new precedent for how U.S. policy intersects with AI deployment. Stakeholders in finance, technology and defense continue to watch closely as the legal landscape evolves.
