The U.S. government just posted a job that basically says "we need someone to decide which AI models are too dangerous to deploy" — and it pays up to $197,200.
NIST’s Center for AI Standards and Innovation (CAISI) opened a 13-month term position for a Member of Technical Staff on its Frontier Assessment team. The job description is a checklist for the AI safety world: evaluate U.S. and foreign systems, track capability levels, and assess threats to national security from cybersecurity, biosecurity, and chemical weapons.
What the Job Actually Demands
This isn’t a policy wonk role. CAISI wants someone who can "develop, adapt, or execute evaluations of AI systems" and "build and maintain experiment infrastructure." That means writing evaluation software, running ML experiments, and producing briefings for key government stakeholders. The role sits in San Francisco or Washington, DC — two hubs that reflect both where the models are built and where the policy decisions get made.
Salary ranges from $121,785 to $197,200 at the ZP-4 to ZP-5 grade levels, with potential to extend the term up to four years. The security clearance required? Sensitive Compartmented Information (SCI). That’s the kind of clearance you need when you’re looking at intelligence-community threat data.
Why This Matters for AI Developers
For anyone shipping frontier models in 2026, this posting signals that the U.S. government is serious about operationalizing AI evaluation. CAISI’s mandate includes tracking "AI diffusion" — that’s government-speak for watching where models and weights travel. The job explicitly mentions agent-based performance in domains relevant to national security. Translation: if your model can autonomously exploit a vulnerability or design a novel toxin, the government will have a dedicated team evaluating that capability.
Congress and executive orders have been talking about AI standards for years. This hire turns talk into people who can actually run evaluations and then brief decision-makers.
The Banned Models Pipeline
That provocative headline — "person who decides which models to ban" — isn’t far off. While the job description uses phrases like "inform understanding of capability levels" and "assess national security risks," the endgame is clear: these evaluations will feed into export controls, deployment licenses, and potentially direct prohibitions. NIST has been the technical backbone for the AI Safety Institute, and this role closes the loop between lab-scale testing and real regulatory decisions.
If you’re building a frontier model, someone on this team will eventually run your system through their gauntlet. The question is whether you’ll be ready for that evaluation.
Source: US feds are actively hiring "person who decides which models to ban"
Domain: usajobs.gov
Comments load interactively on the live page.