Anthropic: U.S. demands unhackable LLMs after Fable 5 release

Anthropic is facing formal scrutiny from U.S. government officials after releasing Fable 5 without waiting for a designated clearinghouse envisioned by President Trump’s cyber executive order, the dispute surfaced on Jun 15, 2026. Government sources accuse the company of ignoring the executive order and releasing the model before voluntary oversight mechanisms were in place.

What happened

Officials told Axios that Anthropic released Fable 5 while the clearinghouse called for in the executive order had not yet been set up. One administration official framed the conflict bluntly: "Everybody said Anthropic was a bad actor. Some of us said it was time to give them a chance. Now those people are questioning that. They screwed us." That official and others also accused Anthropic of proceeding despite knowing a jailbreak risk could exist, saying, "They came to every fork in the road and took the wrong fork."

The tip about a reported jailbreak, whose existence and severity remain unconfirmed, reportedly came from Amazon and other tech companies. The Department of Commerce and Anthropic are in talks, and sources say additional meetings are planned that will involve the CIA and science advisor Michael Kratsios.

Government and industry response

Over 100 security experts and technology executives published an open letter asking Trade Secretary Lutnick and National Cyber Director Cairncross to lift export controls on Anthropic’s models Fable and Mythos. Signatories include Alex Stamos, Rachel Tobac, Katie Moussouris, Dan Lorenc, and Joe Levy.

The letter argues that Anthropic’s models are effective at finding software security flaws, but they are not uniquely capable of that work. It names other models that can perform similar tasks, including GPT-5.5, Opus, Sonnet, and the Chinese model Kimi 2.7. The signatories warn that export controls are depriving defenders of tools while Chinese open-weight models are "only months behind the top U.S. models."

Anthropic also built several safeguards into Fable that much of the security community dismissed at launch as overkill. The debate now centers on whether those measures were sufficient and whether the company should have waited for a formal government clearing process.

Security experts cited in the discussion emphasize that LLM security is an unresolved technical problem. OpenAI has warned that prompt injection, a related attack method, may never be fully solved. Anthropic’s own CEO, Dario Amodei, warned in 2023 that a jailbreak could be severe, saying it could be "life or death" if safety protocols were bypassed in domains like science, technology, and biology.

Why it matters

The disagreement exposes a disconnect between what government officials expect—models that are effectively unhackable before international distribution—and the current technical reality, where prompt injection and similar vulnerabilities have no complete fix. If the U.S. insists on unhackable frontier models as a precondition for international shipment, companies and regulators will face hard tradeoffs between speed of deployment, defensive capabilities, and geopolitical risk management.

The open letter adds another dimension: imposing export controls on U.S. tools could limit defenders’ access to state-of-the-art models while competitors elsewhere close the gap. That would shift the balance between defense and offense in cyber and vulnerability research, according to the signatories.

What to watch

Follow the Commerce-led talks with Anthropic and whether they produce a retroactive signoff or a formal clearinghouse process. Watch for decisions from Trade Secretary Lutnick and National Cyber Director Cairncross on export controls for Fable and Mythos, and track any published technical assessments of the alleged jailbreak tip originated by Amazon and other companies. These signals will show whether the dispute becomes regulatory precedent or a contained policy skirmish.