The U.S. government is in advanced talks with OpenAI, Anthropic, Google, and other major AI labs about voluntary standards for releasing frontier models, with an announcement possible as soon as early July. The framework would set benchmarks for advanced models, release timelines, and clarify who can access them domestically and internationally.

The negotiations signal a turning point in how Washington approaches AI. Instead of banning or heavily restricting models, the administration is trying to shape how they get released. Companies participate voluntarily. In exchange, they get certainty about what the government will ask of them next.

What the Framework Would Include

The centerpiece is voluntary submission of frontier models to the government for up to 30 days of cybersecurity review before public release. Companies share models with vetted reviewers. Reviewers check for vulnerabilities. The company then decides whether to proceed, delay, or modify the release.

The framework also addresses access controls. Who can use a cutting-edge model? How are access levels enforced? What happens if a model leaks or gets misused abroad? These questions matter more as models grow more powerful.

What Companies Are Already Doing

The framework is being operationalized in real time. OpenAI delayed GPT-5.6’s full launch at the government’s request, limiting access to a small group of vetted partners. Anthropic’s export restrictions lasted three weeks before lifting. Google is in talks ahead of releasing advanced coding models.

No company wants to be singled out. Voluntary standards give everyone a playbook. Do this, and the government stays quiet. Ignore it, and you might get the Anthropic treatment.

Why This Matters Now

The government’s leverage is clear. Export restrictions work. Companies move fast when Washington asks nicely. But the long-term relationship matters more than any single model release. A framework lets both sides know what to expect.

For overseas users and companies, this could mean delayed access to new models. For American companies, it means they know what hoops to jump through. For the government, it’s soft power: the ability to influence AI development without legislation.

Voluntary standards are only voluntary if not complying brings consequences. The government just proved it can enforce that, quietly and quickly.