US Government Finalizes Voluntary Framework for AI Model Security Tests

The US government has finalized a voluntary framework for testing advanced AI models' cybersecurity vulnerabilities. This framework, established under Executive Order 14409, allows federal agencies to review frontier AI systems for up to 30 days before they go public. The NSA and CISA are the primary agencies involved in conducting these evaluations.

Industry players had initially pushed back against a proposed 90-day review period, arguing it would slow development cycles and give foreign competitors an advantage. After negotiations with companies like Anthropic, OpenAI, and Google, the government cut the window by two-thirds to 30 days.

The framework's effectiveness depends on voluntary uptake from major AI labs. If participation is consistent, it could build legitimacy and potentially evolve into a more formal security evaluation process.