Fable #6: The Return of the King
Zvi's AI Roundups · Zvi Mowshowitz · 2026-07-03
US export controls on Anthropic's frontier AI models Fable 5 and Mythos, triggered in June 2026 after Amazon researchers flagged a capability concern with a routine code-debugging request, were fully lifted on June 30 after Anthropic tightened safety classifiers that now also block legitimate debugging tasks.
Appears in
Extraction
Topics: ai-export-controlsanthropic-modelsai-safety-policyus-ai-regulationfrontier-model-safety
Claims
- The US government applied export controls to Anthropic's Fable and Mythos models on June 12, 2026, triggered when Amazon researchers showed the model could respond to 'fix this code' requests.
- Export controls on Mythos were lifted June 26 and on Fable fully on June 30, with worldwide access restored July 1.
- Anthropic's new classifiers, required to satisfy the government, now cause routine debugging tasks to fall back to lesser models, because the triggering 'jailbreak' was itself a standard debugging request.
- The entire regulatory episode was handled ad hoc by political actors (Lutnick, Bessent) without a systematic evaluation regime, which CAISI experts had assessed as safe from the start.
- GLM-5.2 is likely the best open-weight model but is substantially behind GPT-5.5 and Opus 4.8 in capability, contrary to widely circulated WSJ reporting.
- OpenAI has formally offered to give the US government a 5% equity stake, which Zvi frames as a likely corrupt shakedown if the stake goes to the government rather than directly to citizens.
Key quotes
The Amazon 'jailbreak' was 'fix this code.' Debugging is literally 'fix this code.' So I don't know what you want Anthropic to do here.
The fiasco continues, at least until such time as we have a systematic regime in place for future frontier models rather than decisions being made ad hoc, by people like Lutnik and Bessent who do not know how any of this works.
This was a huge own goal for the US, and we will see how bad US models get over the next six months and if Chinese models become noticeably better for cyber work.