White House Asks OpenAI to Gate GPT-5.6 Access on a Customer-by-Customer Basis
In brief: The White House's cybersecurity and science offices told OpenAI to release GPT-5.6 only to vetted partners first, with the government approving access "customer by customer" before any wider rollout.
Key facts
- White House ONCD and OSTP requested OpenAI stagger the GPT-5.6 launch over cybersecurity concerns
- CEO Sam Altman told staff the government would approve each customer during a preview window
- Core fear: frontier models can identify and exploit software vulnerabilities at speeds beyond human capability
- If the limited preview goes smoothly, a broader public release is planned "a couple of weeks later"
- Marks the second US government intervention into a frontier model launch — after the Fable 5 export-control directive against Anthropic
Why it matters
The federal government has now directly shaped the release of two consecutive frontier AI models. What began as an export-control exception for Anthropic appears to be hardening into a standing framework: major labs may need government clearance before releasing their most capable models — a significant shift from the open-release norms of the past five years.
Read more
- TechCrunch — TechCrunch
- Axios — Axios