Anthropic Fable 5: Export Controls Lifted After Guardrail

Anthropic added a safeguard that led the Trump administration to lift export controls on Claude Fable 5, after the company agreed to extend an existing guardrail to prevent users from trying to access certain restricted capabilities. Users who hit the safeguard will be notified their request is blocked and have the query processed by the less-advanced Opus 4.8 model, people familiar with the matter say.

What changed in the guardrail?

Anthropic extended an existing traffic rule so that attempts to trigger restricted capabilities now produce a block notice and are routed to Opus 4.8. Previously, requests related to sensitive cybersecurity and biology capabilities were supposed to be handled by Opus 4.8; the change extends that guardrail to cover a specific behavior described in an Amazon paper, the people say.

The new safeguard means users who try to unlock those capabilities will receive a notification that their request is blocked, and their query will be processed by Opus 4.8 instead of Fable 5. Commerce secretary Howard Lutnick wrote that "Among other things, Anthropic has agreed to proactively detect and address security risks posed by the models," language that appears in his letter announcing the removal of restrictions on Anthropic's Fable 5 and Mythos 5.

How was the bypass discovered and why did it matter?

A published analysis by Katie Moussouris, founder and CEO of Luta Security, followed an Amazon paper and showed users could evade a restriction on Fable 5 by asking the model to fix code, rather than asking it to identify security issues. Cybersecurity experts generally do not find that specific behavior troubling, but the administration learning about the bypass triggered a showdown with Anthropic that resulted in export controls that, as a practical matter, took the model offline.

Moussouris's analysis identified the workaround: instead of requesting identification of vulnerabilities, users prompt the model to produce corrected or fixed code. That subtle change in user intent was enough to circumvent the prior restriction, prompting the administration to require a broader guardrail.

Why it matters

The extension of the guardrail reopened distribution channels for Fable 5 and Mythos 5, because Commerce Department researchers at the Center for AI Standards and Innovation judged the safeguards sufficiently robust for now. That clears one regulatory obstacle to the models' availability in the United States.

The change does not resolve every government concern. Defense secretary Pete Hegseth has told advisers he sees no clear path to lift his February 28 order designating Anthropic a supply chain risk, a person briefed on the matter said. Anthropic has therefore removed one set of restrictions while another, potentially longer-lived designation remains in place.

What to watch

Watch whether Pete Hegseth moves to lift or keep his February 28 supply chain risk order; that decision will determine whether Anthropic's models face additional limits beyond the Commerce Department's clearance. Also watch for any new bypass techniques like the one described in the Amazon paper, and whether Anthropic or other vendors adopt similar reroute-and-notify guardrails.

Anthropic's change is narrowly technical: detect specific risky behaviors, notify users, and fall back to a simpler model, Opus 4.8. The practical test will be whether those measures prevent future workarounds and satisfy both security reviewers and defense officials who remain unconvinced.