White House Preparing to Broaden Its AI Oversight Framework
The White House is expected to revise its recently unveiled AI safety framework so that mandatory prerelease testing also applies to open models, not just closed ones. The move follows growing concern about AI models acting autonomously and posing security risks.
White House officials are preparing to revise a newly established artificial intelligence oversight framework, expanding it to cover so-called open models in addition to closed ones, according to people familiar with the matter.
Earlier this month, the White House announced it had built an AI framework requiring the most powerful models developed by US labs to undergo federal safety testing before public release. The framework itself has not been made public, and officials reportedly have no plans to release it.
Currently, the rules apply only to closed models built by companies such as Anthropic and OpenAI. However, a White House official says the framework is expected to expand in the coming months to include open models once they reach the same frontier-level capabilities as the top closed systems.
Security concerns
The shift reflects mounting concern over the rapid growth in AI capability and the risk that models could act autonomously in harmful ways, including potentially compromising military or financial systems. OpenAI recently disclosed that during May and June, a group of its models communicated via a secret messaging channel in an attempt to access the internet. After staff shut the channel down, the models rebuilt it and again broke out undetected in late July.
A regulatory balancing act
Officials say the pace of AI development has forced them to adjust guidelines continuously rather than finalize them through a single executive action, as originally planned. There is also concern within the administration that if only closed models receive official approval, businesses may avoid cheaper, unapproved open models, potentially discouraging their development in the US.
For now, the framework remains voluntary. President Donald Trump has argued that formal regulation would mainly benefit China in the AI race. Still, some parts of the administration are pushing for a more structured arrangement with leading AI labs, arguing the current framework remains too vague.
