OpenAI confirms it will comply with President Trump's EO that asks AI companies to allow the US government to assess their models' capabilities before release
OpenAI has confirmed it will comply with Donald Trump's executive order that asks AI companies to allow the federal government …
Context & Ripple Effects
OpenAI had already backed pre-release government access in 2024, when it and Anthropic agreed to give the US AI Safety Institute early access to major models. A May draft of the White House order pointed toward a voluntary framework for such access.
The company’s more recent policy paper did not reject evaluation itself; it argued that cyber-risk assessments for advanced systems should be mandatory and led by CAISI rather than the NSA. Its confirmation therefore aligns with the access mechanism while preserving a dispute over which agency should run it.
First-order effects
- OpenAI will make its models available for US government capability assessment before release under the executive order’s process, adding a formal federal review step to its launch planning.
- The company gains alignment with the administration’s requested framework, even as its policy position favors CAISI-led cyber evaluations over NSA leadership.
Second-order effects
- Other frontier-model developers face stronger pressure to offer comparable pre-release access, building on the earlier OpenAI-Anthropic arrangement with the US AI Safety Institute.
- The practical policy contest shifts from whether government evaluations occur to their scope, whether they are voluntary or mandatory, and which federal body receives the models and findings.
Third-order effects
- If this practice becomes standard, pre-deployment evaluation could become a durable operating requirement for leading AI developers rather than an ad hoc voluntary commitment.
- The emerging structure may concentrate influence over frontier-model governance in the federal evaluation body selected to administer reviews; OpenAI’s CAISI-versus-NSA stance shows that institutional control remains unsettled.
The trend: US AI policy is moving toward routine government visibility into advanced models before public release, with the central unresolved question being who governs those assessments.