Anthropic and Trump Administration Deadlocked Over Claude Fable 5 Security Review

BLUF

Export restrictions on Anthropic’s Claude Fable 5 remain in place after the Trump administration concluded talks with the company on Monday without a resolution. According to Wired AI, administration officials continue to believe methods exist to circumvent the model’s safety features to access more potent cybersecurity functions. The failed negotiation leaves Anthropic facing simultaneous regulatory pressure from both the executive branch and the Pentagon over AI capabilities and military applications.

Export Controls Hold Despite High-Level Discussions

The Commerce Department and Anthropic’s leadership conducted emergency talks at the Commerce Department this week, attended by a roster of security and AI researchers from the Center for AI Standards and Innovation and the Office of the National Cyber Director. According to Wired AI, Commerce Secretary Howard Lutnick participated remotely from the G7 summit in Evian, France. Anthropic fielded a delegation including cofounder and chief compute officer Tom Brown, head of external affairs Sarah Heck, and security researchers Logan Graham and Nicholas Carlini.

Despite these efforts, the two sides left the table without lifting the restrictions first imposed last week. Wired AI reports that an Anthropic spokesperson stated “both parties are working quickly to get this resolved,” but neither side disclosed conditions for a breakthrough or a timeline for resumption of consumer access to the model.

The NSA’s Role in Triggering the Lockdown

The vulnerability alert originated from Amazon CEO Andy Jassy, who contacted Treasury Secretary Scott Bessent directly last week. According to Wired AI, Jassy’s call “played a role in spooking the administration,” and the White House subsequently tasked the National Security Agency to evaluate the technical threat.

The NSA’s assessment proved decisive: according to Wired AI’s reporting, the agency concluded it was plausible to strip Fable 5’s guardrails, prompting Commerce Secretary Lutnick to draft export-control language. Anthropic maintains that the administration overstates the threat, reiterating this stance during the working-group meetings. The company has voluntarily suspended user access to Fable 5 in the interim.

Anthropic’s Escalating Pentagon Standoff

The Claude Fable 5 dispute arrives as Anthropic navigates a prolonged separate conflict with the Department of Defense over whether its AI models can be deployed for certain military operations. According to Wired AI, this Pentagon tension adds urgency to the current crisis and marks an unusually fraught period for the company’s government relations.

The dual regulatory pressure—one focused on export controls, the other on defense applications—suggests that Anthropic’s risk profile with the federal government has shifted materially in recent weeks.

Why This Matters

For AI safety teams and enterprises relying on Claude Fable 5 for downstream applications, the indefinite restriction creates operational uncertainty. The fact that the Commerce Department expressed openness to lifting controls contingent on a “full resolution” of concerns signals that negotiation remains possible—but the bar for resolution appears high.

The case also establishes a precedent: unconfirmed jailbreak claims, even when reported by a major corporate stakeholder like Amazon, can now trigger rapid export restrictions without public disclosure of the vulnerability’s technical details. This asymmetry of information—the government holds evidence; Anthropic and the public do not—may reshape how AI companies engage with safety disclosures in the future.

For rival AI labs including OpenAI and Google DeepMind, the incident underscores the regulatory risk of models with dual-use cybersecurity capabilities. It also signals that the Trump administration is prepared to use export controls as a tool to enforce its interpretation of AI safety, rather than relying solely on voluntary commitments from vendors.