The Trump administration has reportedly cleared Anthropic to release its Mythos AI model to a limited group of companies and government agencies, partially reversing a broader access cutoff that had also taken Anthropic's Fable 5 model offline. Anthropic had disabled access to both Fable 5 and Mythos 5 after a government export control directive cited national security authorities. The selective restoration draws a hard boundary between approved domestic recipients and the wider market.
How the Cutoff Happened
Anthropic's decision to disable Fable 5 and Mythos 5 was compliance-driven, not commercial. A government directive grounded in national security authorities required the company to pull access — a step that affected any customer base relying on either model. The mechanism here is export control authority, a regulatory framework that has historically governed physical goods: semiconductors, fabrication equipment, dual-use hardware. Applying it to AI model access represents a meaningful extension of that framework, one that positions frontier models as controlled items whose distribution the government can manage directly.
What the Reported Clearance Actually Covers
The Trump administration's reported permission targets two categories: some companies and government agencies. The scope of "some companies" and the identity of specific agencies have not been publicly specified in the available reporting. This is not a full market reopening of Mythos 5 — it is a managed distribution channel with criteria that have not been disclosed publicly.
Fable 5 does not appear in the reported clearance. Whether that model remains fully restricted or is on a separate track is not addressed in current reporting.
The Regulatory Signal
The national security framing of the original directive points to concern over foreign access to Anthropic's models, though the specific threat assessment behind the restriction has not been made public. That the government moved to control access at the model level — rather than at the infrastructure or chip layer — suggests regulators are treating advanced AI capabilities as a distinct category of controlled technology. For Anthropic, navigating that framework now appears to be an ongoing operational reality, not a one-time compliance event.