Share with your CIO
The Federation of American Scientists is pushing the next administration to expand the GSA’s existing AI Center of Excellence into a federal hub for state and local AI procurement, covering the roughly 90,000 state, local, tribal, and territorial agencies currently buying AI tools with little structured guidance. The proposal is a “hub and spoke” model, placing fellows drawn from academia, government, and civil society at the center, then routing their expertise out to agency cohorts navigating vendor contracts, model selection, and deployment governance. The 12-month plan moves from fellow recruitment to a pilot cohort to a scaling strategy.
What this means for your business
If your enterprise sells into government, or competes for talent and contracts against vendors who do, the operative question is whether SLTT agencies are about to get dramatically better at saying no. Right now the asymmetry is stark: Microsoft, Palantir, and their peers field entire public-sector sales teams against procurement officers who may never have evaluated an AI contract before. A center of excellence that trains those officers changes the negotiating dynamic at every renewal table, not just the first deal.
The proposal’s honest vulnerability is timing and political durability. A hub that depends on rotating fellows and voluntary agency participation is structurally fragile. The recurring failure mode in federal technology initiatives looks exactly like this: strong pilot cohorts, positive first-year reviews, then a budget cycle where the program gets absorbed or zeroed out before the knowledge transfers deeply enough to survive staff turnover. The FAS memo, written by researchers with an incentive to see federal investment in responsible AI infrastructure, underweights this institutional decay risk and leans heavily on the Microsoft responsible-AI hub as a comparator, which glosses over the difference between a private company mandating internal compliance and a federal body trying to persuade 90,000 sovereign agencies.
The more durable effect may be regulatory rather than operational. If even a partial version of this center gets stood up, it creates a documentation trail of what “responsible procurement” looks like in practice, and that trail becomes the reference point when liability questions arise. Recent AI regulations are already moving toward shared liability between vendor and deploying agency. Any enterprise AI vendor whose contracts with SLTT agencies don’t reflect that shift is holding paper that will get renegotiated, one way or another. The question to weigh at your next government contract review isn’t whether this center succeeds; it’s whether your contract language was written assuming the old liability model.
Concept deep-dive: Foundation model procurement
A foundation model is a large AI system trained on broad data that can be adapted to many specific tasks, the way a general-purpose operating system runs many different applications. When a government agency buys access to one, say Azure OpenAI, it isn’t just procuring a finished tool but taking on responsibility for how that tool gets configured and deployed. That shifts accountability upstream to the buyer in ways traditional software procurement contracts, written for finished products, were never designed to handle.
Based on reporting from Expanding Local Government Capacity for AI Procurement and Use, originally published 2025-02-14 03:00:00.

