The Trump administration’s new framework for assessing cybersecurity risks from advanced AI explicitly excludes open models from its testing requirements. According to Axios, these voluntary guidelines not only omit open source software which allows anyone to inspect core components but also state that the rules cannot be applied to restrict such models after they have been released. This approach emerged following an executive order signed in June that asked AI companies to share frontier models with the federal government before publication to address security concerns. The directive creates a clear distinction between proprietary systems and open alternatives, effectively limiting the government’s ability to audit the latter before they reach the public.
This limitation matters because it removes a potential avenue for external review of unregulated technology. By refusing to test open models, the administration accepts that these systems will operate without the same level of pre-deployment scrutiny applied to closed systems. The decision prioritises speed of deployment over comprehensive security verification for a large segment of the AI market. It leaves the burden of safety entirely on the developers rather than the state.
- Open source models are legally exempt from pre-release federal testing.
- The framework prohibits using these guidelines to restrict released open models.
- Security reviews are limited to proprietary systems sharing data with the government.
