Anthropic Fable 5 and Mythos 5 return as US lifts export controls

This episode is a vivid illustration of how frontier model deployment is now co‑governed by AI labs and states. In mid‑June, US export controls forced Anthropic to pull Fable 5 and Mythos 5 worldwide over jailbreak‑driven cyber‑risk concerns. Two weeks later, access is coming back—but only after Anthropic stood up new security classifiers, agreed to deeper collaboration with Commerce, and began pushing an industry framework for rating jailbreak severity and coordinating responses.([anthropic.com](https://www.anthropic.com/news/redeploying-fable-5?pubDate=20260206&utm_source=openai))

For the race to AGI, the big story isn’t just that a top-tier model went dark and returned. It’s that export controls are now being aimed at models themselves, not just chips, and that a lab survived a full‑scale access shutdown by working through the new policy stack. That sets a rough playbook for future incidents: temporary clampdown, intensive joint testing, then constrained reopening with new guardrails. Labs that can adapt quickly will have a strategic edge; those that can’t may find their flagship models stuck in regulatory limbo.

Competitively, Anthropic emerges with scars but also credibility: it can say it passed a high‑stakes national‑security review while keeping a premier model online again in weeks, not months. But it also confirms that US regulators are willing to intervene directly in commercial rollouts, which every frontier lab—from OpenAI to Chinese incumbents—now has to treat as a live operational risk.([semafor.com](https://www.semafor.com/article/07/01/2026/us-ends-export-ban-of-anthropics-mythos-and-fable-models))