US government is building an AI licensing regime without actually writing the rules

The US government has quietly started telling AI companies what they can and can’t do with their most powerful models. The catch: there’s no formal law or regulation backing any of it.

On June 13, 2026, the Commerce Department sent a letter to Anthropic requiring that access to the company’s latest AI models be licensed for foreign individuals, including the company’s own domestic employees who happen to be foreign nationals. Commerce Secretary Howard Lutnick’s name was on the directive. Anthropic’s response was swift and blunt: it suspended access to its newest models for all foreign nationals.

A licensing regime that doesn’t technically exist

The federal government hasn’t passed an AI licensing law. There’s no formal regulatory framework. No published guidelines that AI companies can read, interpret, and comply with.

What exists instead is something Leo Schwartz of The Information describes as a de facto licensing regime. The government is wielding national security authority to impose restrictions on a case-by-case basis, without the transparency or consistency that formal rulemaking would require.

Anthropic isn’t the only company getting the treatment. The Trump administration also requested that OpenAI limit the rollout of GPT-5.6, its latest model, pending government approval for each individual customer.

The approach centers almost entirely on national security concerns, with particular focus on foreign AI talent from China. Rather than establishing broad rules that apply equally across the industry, the Commerce Department appears to be making decisions model by model, company by company, letter by letter.

Why this matters for the AI industry

When Anthropic suspended access for all foreign nationals after receiving the Commerce Department’s letter, it wasn’t making a political statement. It was making a risk calculation. Without clear rules about who counts as a security concern, the safest move was to cut off everyone who might fall under the government’s vague mandate.

OpenAI faces a similar bind. Being told to seek government approval for each customer before rolling out GPT-5.6 transforms a product launch into something closer to a defense contracting process.

National security versus innovation, the eternal tension

The Trump administration’s approach represents a sharp departure from earlier AI governance efforts, which leaned heavily on voluntary commitments from companies and generally prioritized encouraging innovation. The pendulum has swung hard toward security-first thinking, and the instrument being used is executive power rather than congressional action.

Discussions are reportedly underway about mandatory licensing requirements and pre-release government reviews of cutting-edge models.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.