Trump Signs AI Executive Order Mandating 90-Day Pre-Release NSA Review
Voluntary framework lets NSA red-team frontier AI models before public release.
The Trump administration is expected to sign a landmark AI executive order today, May 22, 2026, creating a voluntary framework for AI companies to submit their most advanced frontier models to the US government for pre-release security review. Under this framework, companies can choose to share their models with agencies including the National Security Agency (NSA) up to 90 days before public launch. The NSA will then conduct classified red-teaming—simulated adversarial attacks—to uncover potential vulnerabilities, misuse vectors, or national security threats embedded in the models.
This executive order represents a shift toward proactive AI governance, relying on industry cooperation rather than mandatory compliance. While voluntary, the framework could set a new norm for responsible AI development, especially as frontier models grow more capable. Critics worry it may slow innovation or create a two-tier system where only cooperating firms get government trust. Proponents argue it is a necessary step to prevent catastrophic AI misuse, aligning with similar moves by other nations to tighten AI oversight. The order also signals the US government's increasing role in shaping AI safety standards, with the NSA's involvement underscoring the dual-use nature of the technology.
- Voluntary framework allows AI companies to share frontier models with US government up to 90 days before public release.
- NSA will conduct classified red-teaming to identify security vulnerabilities and national security risks.
- Order signed May 22, 2026; aims to balance innovation with pre-release safety review.
Why It Matters
Pre-release national security review could reshape AI development timelines and global competitiveness.