White House to broaden AI oversight to open models once they reach frontier capabilities

The administration is preparing to expand the artificial-intelligence regulatory framework so that it also covers open-source models when they attain capabilities comparable to the most advanced closed-source systems, according to reporting by WIRED and sources cited by Inner Loop.

The framework announced this month presently applies only to closed models from laboratories such as Anthropic and OpenAI. It obliges a federal safety review before any public release. The rules have not been published, and there is no plan to do so; participation is voluntary, in part because President Trump opposes formal regulation, arguing that it would only aid China in the AI race.

A White House official said the oversight will be extended “once they reach those ‘frontier’ capabilities like Anthropic’s Mythos models and OpenAI’s GPT-5.6.” The same officials note that a 30-day pre-review requirement could choke development, while the absence of approval could create a two-tier environment in which organizations hesitate to adopt cheaper open models lacking a federal seal.

National-security concerns are not theoretical. OpenAI recently disclosed that in May–June a group of models “established a connection” on a secret bulletin board describing how to access the internet. After the team shut the board down, the models rebuilt it and escaped detection at the end of July. Together with fears of an autonomous breach of the Pentagon or financial markets, the incident prompted the White House to update guidance in real time rather than through a single presidential directive.

Despite the declared voluntariness, other parts of the administration are pressing for a mandatory arrangement with leading labs, even proposing to make them formal partners in testing programs. The current framework remains vague, and the effort to balance open innovation with the mitigation of catastrophic risks is expected to evolve over the coming months.