Anthropic’s Fable 5, the model that stunned developers and then disappeared in a cloud of government security concerns, could be back as soon as this week. The Trump administration is close to letting the company restore access after a 15-day blackout, a rare case of a top-tier AI model being pulled and then potentially pushed back into circulation by regulators.
The timing matters because the fight over Fable 5 has been more than a one-off spat. It has become a test of whether the U.S. will review frontier AI models through ad hoc intervention or through a clearer process that developers can actually plan around. Anthropic and OpenAI both want the latter; the government has mostly offered the former.
Fable 5’s return is edging closer
Insiders expect the restrictions on Fable 5 could be lifted within days, though the Pentagon and the National Security Agency still need to sign off. That leaves the model in a familiar but awkward state: other agencies appear satisfied, but the final security gatekeepers still hold the keys.
The model’s appeal was obvious from the start. Users praised its deep reasoning and fast coding, while Anthropic positioned it as its most capable public release. One early test, highlighted by the company, said Stripe used Fable 5 to rewrite a 50-million-line codebase in a day – the kind of claim that makes engineers either cheer or reach for the nearest stress ball.
- Fable 5 has been offline for 15 days
- The administration could lift limits as soon as this coming week
- Pentagon and National Security Agency approval is still pending
Mythos 5 got a limited green light first
Friday brought another sign of détente: the Commerce Department allowed Anthropic to restore access to Mythos 5 for a limited number of trusted users. Unlike Fable 5, Mythos 5 was never meant for broad public access; it carries guardrails designed to reduce abuse in cyberattacks or biological terror scenarios.
Commerce Secretary Howard Lutnick said Anthropic had worked with the government to address the risks around both models, and that the company had committed to work on future protocols and standards. That is bureaucratic language with teeth: the administration is signaling it wants voluntary compliance, but also wants a template it can reuse the next time a frontier model scares somebody in Washington.
Developers want access, not drama
The blackout landed hard because Fable 5 was not some research demo hiding in a lab. Anthropic made it available on several paid Claude plans at no extra cost through June 22, then access vanished on June 12, freezing automated work mid-task and sending companies scrambling for substitutes, including cheaper Chinese models. That kind of whiplash is why developers hate opaque model policy almost as much as they hate broken code.
There is also a commercial catch. It is still unclear whether subscribers will get the free access they were promised, or whether Fable 5 comes back behind new fees or identity checks. If Anthropic tightens the screws too far, rivals will happily market themselves as the calmer option, even if they are not the smartest model in the room.
A broader AI review process is still missing
Anthropic and OpenAI are both pushing the administration to formalize how new models get reviewed, rather than handling each case as a one-off. That push lines up with President Trump’s June 2 executive order, which set up a framework for voluntary government vetting of the most powerful new AI systems.
OpenAI’s limited preview of GPT-5.6 on Friday sharpened the complaint. The company argued that government access checks should not become the long-term default because they keep advanced tools away from users, enterprises, cyber defenders, and global partners. For now, though, the government still prefers to negotiate model by model – a process that is exhausting, slow, and very much not built for the pace of AI releases.
The open question is whether Fable 5 returns as a normal product or as a watched one. Given how much attention it already has, ”normal” may be off the table anyway.