230 firms warn GSA AI clause would block open-source models

GSA's proposed AI procurement clause is shaping up as the first real stress test of whether the Trump administration's open-ecosystem AI rhetoric can survive contact with federal contracting machinery.

The rule, drafted as a GSAR deviation for the Multiple Award Schedule, would require contractors to flow down a set of AI safeguarding obligations to every model provider in their supply chain, including obligations around data handling, bias, security incidents, and government-use rights. For closed, first-party models, that's heavy but potentially workable. For open-weight models published by entities with no contractual relationship to the government, it's structurally impossible. There's no one to flow down to.

That asymmetry generated over 75 comments by the August deadline, and the consensus from industry is blunt: the rule as written would exclude open-source AI from GSA vehicles. "Obligations should follow the data, not the model's authorship," Nvidia's Bruce Andrews wrote, capturing the core objection in a single line. The problem isn't the safeguards themselves, it's that the rule assigns responsibility to the wrong party. A platform provider integrating a third-party model can't guarantee that model's neutrality, can't compel intermediate-step disclosure, and can't prevent the model publisher from conducting human content review. "Contractors can only be responsible for matters within their control," the Coalition for Common Sense in Government Procurement noted, and that's exactly what the flowdown framework ignores.

The administration's own letter

The timing couldn't be more awkward for GSA. In July, more than 230 companies (including Meta, Google, OpenAI, Microsoft, Nvidia, and Palantir) signed a letter urging the White House to protect open-weight models, arguing they're foundational to U.S. competitiveness. "Our AI leadership will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector," the letter said. It warned that relying solely on closed models "is not inherently safe" and that open models prevent vendor lock-in.

GSA's proposed clause pulls in the opposite direction. It doesn't ban open-weight models explicitly, it just creates compliance obligations that can't be satisfied by their publishers, making them practically unprocurable through the government's largest commercial purchasing vehicle.

Withdraw or rewrite

Palantir went furthest, formally urging GSA to withdraw the rule entirely, arguing the agency lacks independent authority to promulgate what amounts to a governmentwide AI acquisition standard. But even commenters who didn't call for withdrawal warned of a chilling effect. The Alliance for Digital Innovation cautioned that agencies could be "pushed toward other acquisition pathways, reducing competition and product choice on GSA contracts." Microsoft said the terms "may deter the sale of commercial AI systems to government."

What comes next is an open question. GSA could withdraw the clause, carve out third-party and open-weight models from the flowdown requirement, or issue guidance clarifying which entity bears responsibility. The alternative (pushing ahead as written) would force agencies wanting the most advanced AI capabilities to route around GSA vehicles entirely, the outcome Palantir's Menaka Kalaskar predicted at the July listening session. For contractors and compliance teams watching this space, the comment docket is now the best signal of where the rule is headed.