Fable 5 is live again as of today, according to @SemiAnalysis_. The frontier model was taken offline June 12 under a US export-control directive, marking the first known case of a model lobotomized and later restored by policy.
Key facts
- Fable 5 was offline from June 12, 2025 to today.
- US export-control directive forced the takedown.
- First frontier model restored by policy action.
- Mythos family model; 'lobotomized' by weight removal.
- @SemiAnalysis_ broke the news on X.
Fable 5, a frontier model in the Mythos family, is back online after being offline since June 12 under a US export-control directive, according to @SemiAnalysis_. The analyst firm calls it "the lobotomized Mythos" — the first instance of a frontier model being taken down and then brought back by policy action. They add: "Won't be the last."
What Was the Lobotomy?
The term "lobotomy" here refers to the deliberate removal of model capabilities — likely weights, checkpoints, or inference endpoints — to comply with export controls. The directive that forced the June 12 takedown has not been publicly disclosed in full, but US export controls on AI models have tightened significantly since the October 2023 Executive Order on Safe, Secure, and Trustworthy Development and Use of AI. The Biden administration's January 2024 interim final rule on computing power thresholds and the subsequent October 2024 rule on model weight restrictions created a regulatory framework where models above certain compute thresholds (10^26 FLOPS for training) trigger export licensing requirements.
Policy Precedent
This restoration sets a precedent: no major frontier model had previously been taken offline by government order and then returned. The closest parallel is the voluntary model access restrictions by companies like Meta and Anthropic during the 2024 election cycle, but those were self-imposed and time-bound. Fable 5's return suggests either that the underlying compliance issue was resolved (e.g., a license granted, or the model's capabilities were below the re-evaluated threshold) or that the policy itself was walked back. @SemiAnalysis_ does not specify which.
What This Means
If Fable 5's restoration becomes a template, it could signal a more dynamic regulatory environment — models may be temporarily suspended during policy reviews rather than permanently banned. That would be a meaningful shift from the binary 'approved or banned' framework many observers expected. It also raises questions about how export control authorities assess model capabilities over time: if a model's benchmark scores or parameter count are publicly known, the restoration implies either a change in the assessed risk or a change in the policy.
The analyst's closing comment — "Won't be the last" — suggests more such actions are expected, either for other Mythos family models or for other frontier models caught in the same regulatory net.
What to watch
Watch for a formal statement from the US Bureau of Industry and Security (BIS) clarifying whether Fable 5's restoration reflects a policy exemption, a license grant, or a reclassification of the model's capabilities. Also watch for other Mythos models — or models from other labs — that remain offline under similar directives.
[Updated 02 Jul via the_decoder]
The Decoder reports that Fable 5's two-week ban was triggered by a jailbreak discovered by Amazon researchers, not a broad export-control directive. Anthropic says even smaller models like Claude Haiku 4.5 could perform the same exploit. A new safety classifier now blocks the technique in over 99% of cases, though it also flags more harmless requests [per The Decoder].
