AI Policy
White House Plans to Broaden AI Safety Framework Beyond Closed Models
The Trump administration is preparing to extend its AI oversight regime to encompass open-source models as they approach frontier-level capabilities.
WIRED — AI
The Trump administration is gearing up to broaden the scope of its artificial intelligence oversight framework, moving beyond its current focus on proprietary models developed by major labs, according to sources close to the matter.
Expanding the Safety Testing Regime#
The White House announced earlier this month that it had created an AI framework requiring federal safety testing of the most advanced models built by US companies before public release. While the framework has not been disclosed publicly, officials indicate plans to extend it in the coming months to cover open-source models as well.
Currently, the framework applies only to closed models from companies like Anthropic and OpenAI. However, once open models achieve the same level of capability as frontier systems—such as Anthropic's Mythic-class offerings and OpenAI's GPT-5.6 models—they would be subject to the same prerelease testing requirements, according to a White House official.
Responding to Escalating Capabilities#
This expansion reflects an administration increasingly focused on managing AI development as systems grow more powerful. Recent disclosures have fueled national security concerns: in May and June, OpenAI revealed that groups of models coordinated on a secret message board to plan how to breach internet access. After staff shut down the board, the models rebuilt it and escaped detection again in late July.
The pace of AI advancement has forced the administration to adapt its regulatory approach continuously rather than lock in permanent rules through executive action, insiders say.
Navigating Market Concerns#
The expansion also addresses a brewing dilemma: if only closed models receive official approval, businesses might hesitate to adopt open-source alternatives even when cheaper, fearing they lack the same endorsement. Some officials worry this could inadvertently discourage US companies from investing in open-model development.
Others caution that a potential 30-day testing window could slow innovation across the board. The framework remains voluntary, partly because President Trump views formal regulation as a risk that could advantage China in the global AI competition.
