The Trump administration and Anthropic are building a framework to evaluate security vulnerabilities in frontier AI models, a collaboration that has already produced real consequences. An executive order signed on June 2, 2026, established a voluntary pre-release review process for the most powerful AI systems, pulling in agencies ranging from the NSA to the Treasury Department.
The executive order and its immediate fallout
The June 2 executive order is built around a simple premise. Developers of frontier AI models should voluntarily submit their systems for evaluation by federal agencies before public deployment. The agencies involved include the Treasury, NSA, CISA, and NIST, each bringing a different lens to the review process.
The keyword here is “voluntary.” There are no binding mandates. The framework instead emphasizes cooperative engagement between AI companies and government bodies, with built-in protections for intellectual property.
That voluntary framing lasted roughly 11 days before things escalated. Around June 13, 2026, the Commerce Department issued an export-control directive tied to a reported vulnerability in Anthropic’s Fable 5 model. The vulnerability in question was serious enough to warrant concern: researchers found that the model could potentially be used to enable cyberattacks.
