Trump administration's AI security testing plan excludes open models

The Trump administration's voluntary framework for testing advanced AI models for cybersecurity risk explicitly excludes open-weight models and cannot be used to restrict them after release, according to Axios. The framework followed a June executive order asking AI companies to share frontier models with the government before release.

Why it matters: Exempting open models from pre-release testing while leaving them unrestricted afterward opens a policy gap just as open-weight models are closing in on frontier capability with safety practices that lag behind, a concern raised in recent industry reporting. It signals that near-term US government AI oversight will center on a handful of closed frontier labs rather than the broader open-weight ecosystem.