Fable 5 jailbreak wasn't unique: it worked on eight frontier models

The US used export controls to take Claude Fable 5 and Mythos 5 offline on 12 June 2026, after Amazon researchers found a way past Fable 5's safeguards to identify software vulnerabilities — in one case producing demonstration exploit code. The models stayed dark for more than two weeks. During the standoff, cybersecurity experts wrote an open letter arguing the flaw was not unique to Anthropic, and Anthropic's subsequent investigation confirmed it: the same bypass worked on Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, 4.7 and 4.8, plus GPT-5.4, GPT-5.5 and Kimi K2.7.

The eight-model list came from Anthropic's investigation reported in early July, not from anything known on 27 June — at first, experts only argued the flaw was shared. The point stands: if a jailbreak spans the whole frontier, switching off one lab fixes little, and open-source advocates warn it simply pushes the same capability to cheaper open-weight models that face no export order.

To restore its models, Anthropic trained a new classifier to block the technique, published a draft 'AI jailbreak severity framework', and opened a HackerOne programme for cyber-jailbreak reports. The episode is feeding Washington's push towards voluntary pre-release testing standards for frontier models — the shared-defence approach the industry-wide flaw seems to demand.