U.S. Government Secures Staggered, Approved Rollouts of Frontier AI Models

Overview

• On Friday, OpenAI began a limited preview of GPT-5.6 for roughly 20 government-approved partners with one access route reported through Amazon Bedrock after U.S. officials asked for a staggered release.
• Anthropic is negotiating with Commerce Secretary Howard Lutnick to lift export controls on Mythos 5 and Fable 5 and Washington has reauthorized conditional, U.S.-only access for a small group of cyber defenders.
• The Commerce Department's restrictions followed a detected possibility of a jailbreak in Fable 5 that officials said could let users bypass safety limits and perform harmful cyber or biological tasks.
• Sam Altman told OpenAI staff to work with government safety guidance even when the company disagrees, and OpenAI says it has strengthened protections in GPT-5.6 for sensitive cybersecurity and bio-related requests.
• The new government interventions are reshaping business plans and market timing as OpenAI weighs delaying its IPO to 2027 and the industry faces greater uncertainty about how and when frontier models will reach users.