White House reportedly missed August AI safety-guidance deadline amid Anthropic Mythos dispute

The White House reportedly missed its August 2026 deadline to publish public guidance for AI safety reviews, the latest turn in federal oversight of Anthropic’s Mythos and Fable models. President Donald Trump had signed a voluntary framework in June; a later export-control directive led Anthropic to disable access to both models globally.

The White House reportedly missed its August deadline

The missed deadline concerned public guidance on AI safety reviews. The policy sequence also included a voluntary federal framework and an export-control directive affecting access to Fable 5 and Mythos 5.

The June order set a voluntary framework

President Donald Trump signed Executive Order 14409 on June 2, 2026. It directed agencies to design a voluntary framework for covered frontier models, including a process that could provide the government with up to 30 days of access before a model was released to other trusted partners. The order does not authorize mandatory federal licensing, preclearance or permitting for model development or release.

The order also directed the government to develop classified benchmarks for cyber capabilities. Separately, the UK AI Security Institute evaluated Claude Mythos Preview in simulated cyber tasks. It reported a 73% success rate on expert-level capture-the-flag challenges and said the model completed a 32-step simulated corporate-network attack in three of 10 attempts, averaging 22 steps. The institute estimated a human would take 20 hours to complete that simulation. Its test network had no active defenders or common defensive tools.

Fable and Mythos had different access rules

Anthropic said Fable 5 and Mythos 5 shared the same underlying model but had different safeguards. Fable 5 was intended for general use; Mythos 5 was limited to trusted Project Glasswing partners for defensive cybersecurity. Anthropic said Fable 5 was released for general use on June 9, while Mythos 5 remained partner-restricted.

Anthropic said the U.S. government issued an export-control directive on June 12 that restricted foreign-national access to both models, including access by Anthropic employees. The company disabled Fable 5 and Mythos 5 for all customers because it could not reliably verify users’ nationality in real time. The directive’s scope and Anthropic’s global shutdown were distinct: the company said it disabled both models worldwide to comply.

The bypass and the safeguard response

Anthropic acknowledged a specific Fable bypass—a way to get a model around a safeguard—but described the finding as narrow and non-universal. The company disputed that it demonstrated a distinctive cybersecurity capability in Mythos.

Anthropic said it worked with government evaluators on updated safeguards and trained an improved safety classifier, an automated filter for requests. In the company’s tests, the classifier blocked the specific reported technique in more than 99% of cases. Anthropic also said it sometimes flagged benign coding and debugging requests.

Anthropic restored access in stages

Anthropic reported different restoration dates for the two models. Fable 5 had been offline for 19 days when global access returned, the company said.