The US Government Just Forced Anthropic to Kill Its Newest AI — Three Days After Launch

June 13, 2026 . 5 min read . #AI #Anthropic #Claude

Anthropic spent months building hype around Claude Fable 5. It launched on June 9, was crowned “state-of-the-art on nearly all tested benchmarks,” and was handed out free to every Pro, Max, Team, and Enterprise user.

Three days later, it was gone.

On June 12, the US Commerce Department ordered Anthropic to cut off all access to Fable 5 and its more powerful sibling, Mythos 5, for “any foreign national” — anywhere in the world, including Anthropic’s own foreign employees. Because there’s no easy way to check a user’s citizenship mid-chat, Anthropic’s only option was to switch the models off for everyone.

What Fable 5 Actually Was

Fable 5 was the public-facing version of Mythos, a model family Anthropic had been quietly testing for months. Mythos had already made waves in the cybersecurity world for its “superhuman” ability to find and exploit software vulnerabilities — capable enough that Anthropic kept it internal while it figured out the safety implications.

Fable 5 was the toned-down release version. Anthropic said its capabilities “exceed those of any model we’ve ever made generally available,” with standout results in software engineering, knowledge work, vision, and scientific research. To keep the riskiest parts contained, the model had hard guardrails: any prompt touching cybersecurity, biology, chemistry, or weapons-relevant distillation would get blocked, and the request would silently fall back to the older Claude Opus 4.8 instead.

Pricing was steep for a “free” launch model: $10 per million input tokens and $50 per million output tokens once the free access window — originally set to run until June 22 — ended.

It never made it that far.

Why the Government Stepped In

According to Anthropic, the directive came straight from Commerce Secretary Howard Lutnick’s office, drafted with help from the department’s Bureau of Industry and Security — the same office that handles export controls on sensitive technology like advanced semiconductors.

The trigger was a jailbreak. The government told Anthropic it had become aware of a technique that could bypass Fable 5’s safeguards by instructing the model to read through a specific codebase and identify software flaws in it. Anthropic reviewed a demonstration of the technique and confirmed it worked — but only well enough to surface “a small number of previously known, minor vulnerabilities.” No one, Anthropic says, has found a universal jailbreak that broadly defeats the model’s protections.

Here’s the twist that’s getting the most attention: Anthropic says the exact same jailbreak technique also works on other publicly available models — including OpenAI’s GPT-5.5 — which aren’t subject to any comparable export restriction. In other words, the underlying weakness isn’t unique to Fable 5. It’s just that Fable 5 is the one the government decided to act on.

The Scope of the Shutdown

The order wasn’t narrowly targeted. “Any foreign national, whether inside or outside the United States” includes Anthropic’s own staff — meaning even employees at the company building the model lost access to it. Faced with no practical way to enforce a citizenship-based access wall inside a chat product, Anthropic disabled Fable 5 and Mythos 5 entirely, for every user, everywhere.

Importantly, Anthropic was quick to clarify that none of its other models — Claude Opus 4.8, Sonnet, and the rest of the lineup — are affected. This is, for now, isolated to the Fable and Mythos family.

Adding to the awkwardness: just days before this all happened, The Register had already reported that Fable 5 was being too cautious — refusing innocuous prompts and tripping its own safety filters on harmless requests. So in the span of a week, Fable 5 went from “annoyingly over-cautious” to “too dangerous to exist,” depending on who you ask.

What Happens Next

Anthropic says it’s working to restore access “as soon as possible” and promised to share more technical detail on the jailbreak within 24 hours of its statement. For now, there’s no timeline for when — or whether — Fable 5 and Mythos 5 come back in their original form.

The bigger story here isn’t really about one model. It’s the first time a US government agency has used export-control authority to force an AI company to pull a publicly released chatbot off the market, days after launch, over a vulnerability that — by Anthropic’s own account — isn’t unique to that model at all.

If the same jailbreak works on GPT-5.5 and other frontier models, does pulling Fable 5 actually fix anything? Or did the government just pick the easiest target?

Sources: CNBC, 9to5Mac, NBC News, Anthropic, The Register