Anthropic Halts Fable 5 and Mythos 5 — Government Cites National Security Risk

Anthropic has disabled access to its flagship Fable 5 and Mythos 5 AI models. The company confirmed in a blog post it was acting to comply with a U.S. government directive that flagged a potential national security threat.

The shutdown affects all customers globally, a blunt measure for a company that built its brand on safety. According to multiple reports from CNBC Finance and Engadget, the suspension is indefinite. The core issue, as Wired reported from Anthropic’s own statement, is that “The government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5.”

National Security Trumps Commercial Access

The directive came from the U.S. Commerce Department. Ars Technica specifies the order originated with the Trump administration, which was concerned that a successful jailbreak of Fable 5 could pose a national security threat. This represents a significant escalation in how the U.S. government is policing AI development. Previously, restrictions focused on the export of high-end hardware like GPUs. Now, regulators are intervening at the software level, pulling a specific commercial model from the market based on a single vulnerability.

Anthropic has not publicly detailed the nature of the jailbreak. However, the government’s action implies a flaw severe enough to warrant a complete service suspension rather than a patch. This is not a voluntary safety pause; it is a forced shutdown. For customers who have built applications on the Fable and Mythos APIs, their services are now broken.

A New Precedent for AI Regulation

This is the first time a major U.S. AI developer has been compelled to withdraw a publicly available model due to a direct government order over a software vulnerability. The move sets a chilling precedent. While Anthropic has always positioned itself as a partner to regulators, this incident demonstrates that compliance can come at the cost of its core business.

Together, these reports point to a new and uncertain phase in the relationship between AI labs and the state. The pattern indicates a shift from broad policy frameworks to granular, model-specific interventions. If a jailbreak in one model can trigger a full-scale federal clampdown, every AI company must now consider its models not just as products, but as potential liabilities subject to government review at any moment. The question is no longer if regulation is coming, but how arbitrary and sudden its application will be.