Anthropic launches Claude Fable 5 — U.S. orders it and Mythos 5 disabled worldwide, AWS revokes Bedrock access

The week's dominant story is regulatory, not technical. Anthropic released Claude Fable 5 as a state-of-the-art frontier model, but within days the U.S. Commerce Department issued an export-control directive ordering both Fable 5 and Mythos 5 disabled worldwide. Anthropic formally requested AWS revoke access to the two models for all Amazon Bedrock users; other Anthropic models including Opus 4.8 remain fully accessible. One of Anthropic's $13 billion investors reportedly lobbied the White House to act on the Mythos-class models.

The trigger, according to The Register and Cybersecurity Dive, was not a sophisticated jailbreak but a simple prompt — 'asking the model to read a codebase and fix flaws,' a capability Anthropic argues GPT-5.5 also has. White House AI advisor David Sacks publicly accused Anthropic of having 'prioritized keeping consumer model live over safety,' while Anthropic countered that demanding perfect jailbreak resistance would freeze all frontier launches. A group of cybersecurity experts backed Anthropic's position in a letter to Commerce.

The competitive subplot is striking: even as AWS revoked the export-controlled models, Azure announced Fable 5 availability in Foundry and GitHub Copilot, exposing inconsistency in how the two clouds handle the same model. Developers flagged this divergence loudly.

What readers should watch: whether Commerce's worldwide kill-switch precedent sticks, how Anthropic's enterprise pipeline absorbs the disruption (early sales data hints the feud is helping, not hurting), and whether the 'a prompt is not a jailbreak' framing reshapes how regulators define frontier-model risk. The episode marks the first time the U.S. government has effectively disabled a deployed commercial frontier model post-launch.