The General Services Administration is considering an acquisition rule that might exclude open-source and third-party artificial intelligence, a move that industry tech leaders told the agency could cause a chilling effect on AI procurement.
Despite revisions from an earlier draft and a listening session with interested parties, the over 75 comments submitted by the deadline this month show the GSA’s proposed rule is still not up to stakeholders’ standards.
Tech leaders like Palantir, Nvidia and Microsoft were particularly concerned about the future of — and their role in servicing — third-party, open-source large language models and weights under the rule.
It puts much of the regulatory responsibility on the wrong entity, commenters said — either on the contractor who may not have any ability to control the open or third-party AI as a platform provider, or conversely, on the developer who may not have any access to government data.
Nvidia pointed to a July letter — which it co-signed with more than 230 companies, including Palantir, Microsoft, Meta, Google and OpenAI — urging the protection of open-weight models, likening it to the rise of open-source software that has flourished in the modern computing era.
“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. “Open weight models—AI models that anyone can download, inspect, modify, and run on their own infrastructure—are an important part of that foundation because they make advanced AI more accessible, adaptable, and widely available.”
The proposed rule’s current requirements would put the onus on the open or third-party publisher who might never receive or process government data, “the opposite of the safeguarding outcome GSA intends, and contrary to the national interest,” wrote Bruce Andrews, Nvidia’s chief external affairs officer.
On the contrary, Palantir said it is “a fundamentally flawed and irrational assumption” that an AI platform provider like itself would be held to the rule’s standards for models it has no control over.
Palantir went as far as to formally urge the GSA to withdraw the rule completely, claiming the agency has no independent authority to promulgate a governmentwide rule, in addition to arguing that the clause is anti-commercial and contrary to the Trump administration’s priorities.
“GSA’s Proposed Rule would significantly undermine the value of GSA contract vehicles because the Proposed AI Clause imposes obligations that are fundamentally incompatible with customary practices in the commercial market for AI technology,” the comment said.
Who is responsible?
Two problems emerge with open-source or third-party AI: The rule’s flowdown requirements could apply to a publisher who has no relationship with the government and cannot comply with its obligations, or the requirements would be put upon a platform provider who has no role in the AI.
As written, because the rule’s flowdown requirements could not be fulfilled by open publishers of AI, it would exclude them from government contracting, Nvidia said.
The letter from the over 230 companies warned that relying solely on closed models “is not inherently safe,” as they can be breached, misused or fail in ways outsiders cannot detect. Also, open models and weights help users avoid lock-in to a single provider or “lose the knowledge and capabilities they build over time,” it said.
White House officials said earlier this month that the federal government will only be reviewing closed models for potential security risks and not open ones, though that could change as tech advances, The New York Times reported.
For others, the rule’s requirements of a platform provider for third-party AI are misguided. “Holding a commercial platform provider contractually responsible for every aspect of how a third party’s probabilistic AI system functions once deployed is irrational and contrary to established procurement law,” Palantir’s comment said.
The Coalition for Common Sense in Government Procurement said the “overly broad language” in the clause may have a “chilling effect on contractor use and adoption of AI,” saying their members have concerns about the practicality of contractor implementation of the ”burdensome requirements,” like assuming a contractor owns an LLM instead of operating it.
“Contractors can only be responsible for matters within their control, and contractors operating in these roles cannot as a general matter guarantee model neutrality, prevent commercial providers from conducting human content review, or compel the level of intermediate-step disclosure the proposed clause requires,” the coalition said.
With this confusion and vague definitions like “unbiased AI principles,” companies are warning they may have to pursue government contracting outside of the GSA.
“We continue to be concerned that the proposed terms will significantly alter existing commercial procurement frameworks under which our company currently makes AI technologies available to the government, will require re-engineering and expenses among commercial AI system providers, and may deter the sale of commercial AI systems to government,” Microsoft said.
Put simply, “obligations should follow the data, not the model’s authorship,” Nvidia’s Andrews wrote.
‘Unbiased AI principles’
GSA has made some changes to the part of the rule covering “unbiased AI principles,” which it defines as having LLMs prioritize “historical accuracy, scientific inquiry, and objectivity,” while acknowledging uncertainty and being neutral and nonpartisan. But those changes don’t appear to have assuaged industry concerns.
“To the extent a government official determines that a commercial AI model does not satisfy the ‘unbiased AI principles,’ the Proposed AI Clause would allocate all risk to Palantir to resolve whatever concerns the government may have,” Palantir said.
Several data privacy organizations — including the Center for Democracy and Technology, Electronic Privacy Information Center, the Electronic Frontier Foundation and Upturn — said that while the updated draft better defines “unbiased AI principles,” it needs to be removed altogether.
The groups said the requirements “are not technically feasible, will reduce the accuracy and reliability of AI systems procured by the federal government, and are ideologically driven such that they could be used to target specific political viewpoints by Administrations of either party.”
“While it is a welcome improvement that the new version no longer calls out specific political viewpoints by removing specific mention of ‘Diversity, Equity, and Inclusion,’ the updated draft still retains many of the most concerning requirements from the prior version that pose significant risks to free expression, civil liberties, and model performance and accuracy,” they said.
But without more changes, several commenters cautioned that the future of GSA contracting could be in danger.
“Where those requirements cannot be reconciled with a provider’s standard commercial offering, providers may be unable or unwilling to make their most capable AI-enabled solutions available through GSA vehicles,” the Alliance for Digital Innovation said.
“Agencies could then be pushed toward other acquisition pathways, reducing competition and product choice on GSA contracts and weakening GSA’s role as a centralized source for commercial technology,” the group added.
Read all the comments below: