First the government shut down Anthropic’s Fable 5. Now it’s asking OpenAI to stagger GPT-5.6’s release over security concerns, with the government approving access customer by customer during a limited preview period. Both frontier AI labs are now operating under government sign-off for model releases.
Here’s what happened and why it matters.
What The Information Reported About GPT-5.6
The Trump administration asked OpenAI to stagger the release of GPT-5.6 over security concerns, according to a report from The Information published Thursday. The report cites a memo from OpenAI to its staff.
CEO Sam Altman informed employees during a Q&A session that the company would release GPT-5.6 in a limited preview to a small group of partners. Altman told staff the government would be approving access customer by customer during this preview period, according to the report.
The request came from conversations with two key government agencies: the Office of the National Cyber Director and the Office of Science and Technology Policy.
The June 2 Executive Order Laid the Groundwork
This doesn’t come out of nowhere. On June 2, President Trump signed an executive order directing federal agencies to establish a framework for “secure deployment of frontier artificial intelligence models.”
The order explicitly includes a process for AI developers to voluntarily give the government early access to frontier models up to 30 days before release. The word “voluntary” was doing a lot of work in the original text. In practice, it’s looking less optional by the day.
The Pattern Is Hard to Miss
Two weeks ago, on June 12, the US government effectively killed Anthropic’s Fable 5, the lab’s most capable model. That story broke wide open when reports revealed Amazon CEO Andy Jassy had personally triggered the crackdown.
Now, 13 days later, the same administration is asking OpenAI to gate GPT-5.6 behind customer-by-customer government approval.
The pattern is clear:
- June 2: Executive order for “voluntary” pre-release review of frontier models
- June 12: Government shuts down Anthropic’s Fable 5
- June 25: Government asks OpenAI to stagger GPT-5.6 with customer-by-customer approval
Both frontier labs now operate under government model-release approval. The permission layer isn’t Anthropic-specific anymore. It’s industry-wide.
What This Means for OpenAI Users
If you use OpenAI’s models, this matters in a few ways.
First, GPT-5.6 won’t be a wide launch. It’s starting as a limited preview to a small group of partners. That’s a dramatically different rollout than the GPT-5.5 Instant update that shipped to all paid users earlier this week.
Second, the government is approving customers individually. That means who gets access depends on what the administration considers acceptable risk, not just who signed up first or who pays the most.
Third, this sets a precedent. If Anthropic and OpenAI are both gated, every frontier lab is on notice. Google DeepMind, Meta, and anyone training a model above a certain capability threshold should expect similar scrutiny.
What the June 2 Executive Order Actually Says
The full text of the executive order establishes a “Framework for Secure Deployment of Frontier Artificial Intelligence Models.” Key provisions include:
- Developers must provide the government with results of safety evaluations before releasing a covered model
- The government gets up to 30 days of early access for testing and evaluation
- The framework applies to “frontier models” defined by training compute thresholds, capability benchmarks, or both
- Agencies are directed to establish procedures for “enforcing compliance with safety and security requirements”
The framework was described as voluntary. But when the government asks and both labs comply, the line between voluntary and mandatory gets thin.
Bottom Line
The age of move-fast-and-deploy-everywhere is over for frontier AI labs.
Anthropic built its brand on safety. OpenAI built its brand on moving fast. Both are now gated by the same government process. The competitive advantage isn’t just capability or speed anymore — it’s who the government trusts more.
OpenAI is also reportedly leaning toward waiting until next year for its IPO, according to the New York Times. These two stories together paint a picture of an OpenAI that’s navigating a much more constrained operating environment than it faced even a month ago.
I’ll update this article as more details about the GPT-5.6 preview become available, including exactly who gets access and when.
