Ten agencies. One coordination structure. The numbers matter less than what they represent.
The CAISI TRAINS taskforce, Testing Risks of AI for National Security, has expanded to include more than 10 federal agencies as of May 2026, according to NIST/CAISI. Participating agencies reportedly include the Department of Defense, Department of Energy, and National Institutes of Health, among others. Those names are single-source, drawn from the NIST CAISI page, and should be read as “reportedly include” rather than confirmed roster entries until independently verified.
CAISI operates under NIST and manages the federal framework for evaluating advanced AI models before deployment in sensitive contexts. The TRAINS taskforce is its dedicated instrument for national security risk assessment, with a focus on Chemical, Biological, Radiological, and Nuclear (CBRN) risks, cybersecurity, and critical infrastructure. That focus was present from TRAINS’ origin; the 10-agency milestone signals that the organizational scaffolding around that mission has grown considerably.
The mechanism connecting agencies to models is the updated memoranda of understanding that OpenAI and Anthropic signed with CAISI on or around May 4, 2026, consistent with prior TJS coverage of the five-lab expansion. Per CAISI, the taskforce uses pre-release access to advanced models pursuant to these updated agreements. That access-to-evaluation linkage is a vendor-attributed claim, CAISI and the labs frame it as cooperative; the independent enforcement architecture around it is still developing.
The real question is what 10 agencies actually means for the future of mandatory review. Prior coverage in this pipeline has documented CAISI’s evolution from a voluntary cooperation framework toward something structurally more durable, the voluntary era is ending, as TJS covered on May 9. The agency count milestone doesn’t create mandatory review by itself. But a taskforce that spans DOD, DOE, NIH, and seven or more additional agencies is harder to characterize as a niche interagency pilot. It looks more like infrastructure.
What to Watch
Don’t expect a formal mandatory review announcement to follow immediately. The current CAISI framework is still built on voluntary MOUs. What the 10-agency expansion does is demonstrate that the coordination capacity for mandatory review exists, the agencies are present, the model access agreements are in place, the CBRN evaluation focus is documented. Legislative or executive action converting that capacity into mandatory obligation is a separate step. But the infrastructure for that step is now materially more complete.
The catch for labs and federal contractors is straightforward: TRAINS now touches more of the federal government than it did six months ago. An evaluation finding that flags a capability concern travels further inside the government. That isn’t a compliance requirement today. It’s a reputational and operational exposure that CBRN-adjacent capability development needs to account for in risk planning.