Our Take
OpenAI negotiated a softer constraint than Anthropic received, but the Trump administration is now directly controlling who gets access to frontier models, not just vetting them.
Why it matters
This signals a shift from pre-release security review to active gating of model access by federal officials. Companies shipping advanced AI now face direct approval workflows, not just compliance checkpoints. That changes deployment timelines and customer access strategies.
Do this week
Enterprise buyers: Confirm with your OpenAI contact whether GPT-5.6 preview access requires government approval and what timeline to expect before planning integrations.
OpenAI delays GPT-5.6 after Trump administration request
OpenAI CEO Sam Altman told employees Wednesday that the company would release GPT-5.6 in limited preview form, granting access only to a small group of enterprise customers, in compliance with a request from the Trump administration. The company cited security concerns raised by the federal government.
During the preview period, the Trump administration would reportedly approve individual customer access on a case-by-case basis before they gain the ability to use the model. This represents a more favorable arrangement than the one imposed on Anthropic, which received an ultimatum earlier this month requiring it to suspend access to its Mythos 5 and Fable 5 models. The administration issued an export control directive prohibiting "foreign nationals," including non-US employees of Anthropic itself, from accessing those systems (per The Verge).
The uneven treatment across companies raises questions about the criteria the Trump administration is using to evaluate frontier model releases. OpenAI's approach involves staggered access with federal approval at each stage. Anthropic faced outright suspension with broad restrictions on who can interact with the technology.
Federal approval is now part of the deployment workflow
This is no longer a pre-release security review. The Trump administration is inserting itself into the access-control layer of frontier AI products, deciding which customers get to use which models and when.
That changes the operational reality for companies shipping advanced AI. A company no longer controls its own customer onboarding or rollout timeline. Federal officials now have veto power over deployment decisions on a per-customer basis. This creates unpredictability for both vendors and enterprises planning to adopt frontier models.
The different outcomes for OpenAI and Anthropic also suggest the administration is not applying uniform rules. Negotiation, company size, or relationship dynamics appear to matter. That ambiguity is costly for everyone building on top of these models.
What enterprises should do
If you have pending requests for GPT-5.6 access, expect delays and approval friction. Ask OpenAI directly what criteria the Trump administration is using and what the approval timeline looks like. Do not assume you will gain access on your original schedule.
For teams planning integrations, lock your architecture to work with current-generation models (GPT-4 Turbo, Claude 3.5 Sonnet) until GPT-5.6 access is confirmed. Building a hard dependency on a model in preview and subject to federal approval is a liability.
Document your use case and customer location now. Non-US entities and foreign-national team members may face access restrictions. This is especially critical for multinational teams or companies with international customers.