OpenAI diverges from Trump's AI EO in a new policy paper, proposing cyber risk evaluations for advanced AI systems be mandatory and led by CAISI, not the NSA
OpenAI's new proposal comes as its CEO Sam Altman descends on Washington for a series of Wednesday meetings with White House officials …
Context & Ripple Effects
OpenAI has been building a more active Washington policy presence, including expanded lobbying and earlier White House recommendations warning that fragmented state legislation could impede U.S. AI progress. Its recent state-level “reverse federalism” effort also shows it is pursuing multiple routes to shape safety rules.
This paper comes alongside OpenAI’s stated willingness to comply with the administration’s pre-release model-assessment request. The divergence is therefore over the design and institutional ownership of cyber-risk oversight, rather than a rejection of federal evaluation altogether.
First-order effects
- OpenAI is placing a concrete alternative before White House officials: mandatory cyber-risk evaluations for advanced systems, administered by CAISI rather than the NSA.
- The proposal publicly differentiates OpenAI’s preferred oversight model from the administration’s EO approach while preserving the company’s broader posture of engagement with federal assessment requirements.
Second-order effects
- Other frontier-model developers and AI policy groups may have to state whether they support mandatory cyber evaluations and which agency should oversee them, making institutional control a central policy dispute.
- CAISI gains visibility as a potential civilian-facing evaluator, while an NSA-led role becomes more contested by companies seeking to influence how sensitive model testing is governed.
Third-order effects
- If federal pre-release evaluation becomes established practice, AI governance could shift from broad voluntary safety commitments toward standardized, agency-run testing for the most capable models.
- The continuing split between federal action and OpenAI’s state-level safety strategy suggests that the eventual U.S. regime may be shaped as much by competition over jurisdiction and implementation as by agreement on safety goals.
The trend: Frontier AI firms are moving from general calls for regulation toward competing proposals over the mandatory tests, agencies, and jurisdiction that would operationalize AI oversight.