Anthropic and OpenAI partner with Trump administration on AI model evaluation framework
In brief
- Anthropic and OpenAI are finalizing a federal government plan for AI model evaluations with the Trump administration
- Frontier AI models must be reviewed for cybersecurity and national-security risks before public release
- A classified benchmarking framework allows developers to grant government access to advanced models up to 30 days before release
- Both companies adjusted their existing Commerce Department pre-release review agreements to align with the new AI Action Plan
Framework and Benchmarking Process
Anthropic and OpenAI are reportedly working closely with the Trump administration to finalize the plan. The executive order mandates the creation of a classified benchmarking process and a framework allowing developers to grant government access to advanced models up to 30 days before release. This gives federal agencies a window to assess frontier systems for potential security vulnerabilities before they reach the public.
Both companies have existing pre-release review agreements with the Commerce Department, and these agreements have been adjusted to align with the new AI Action Plan. The move represents a formal institutionalization of what had previously been informal consultations between private AI labs and government regulators.
Structured Oversight and Timeline
This cooperation suggests a shift from informal consultations to a more structured evaluation process. The arrangement gives the Trump administration direct visibility into advanced AI capabilities while allowing developers to maintain competitive timelines for model releases.
The next few days are critical as the July 31 deadline approaches. How the framework is finalized could set a precedent for how government agencies evaluate and oversee frontier AI development going forward.
