Key Takeaways

  • GSA’s revised AI contract clause narrows licensing terms and recognizes contractor ownership of underlying models.
  • The Professional Services Council still sees concerns involving scope, subcontractor requirements, data rights, and intellectual property.
  • Cyberattacks on water systems add urgency to calls for faster federal acquisition and restored prototyping authorities.

The General Services Administration has revised its proposed artificial intelligence contract clause after contractors raised concerns about intellectual property, licensing, and an unusually broad definition of AI.

The Professional Services Council president described the new draft as “light years better” than the version circulated in March. Comments on the revision were due August 3, following a GSA listening session in mid-July. For contractors watching the government’s approach to AI procurement, the changes suggest that industry feedback is shaping the policy before it becomes a final contract requirement.

Among the key revisions, GSA narrowed the proposed license grant to the federal government and clarified that contractors would retain ownership of underlying large language models and base models. The agency also replaced a limitation requiring development and production in the United States with a preference for domestic work.

These modifications directly affect commercial AI developers. Retaining rights to foundational models can affect whether a contractor is willing to offer its technology to the government at all, particularly when the same model supports multiple commercial and public-sector customers. Broad government rights could discourage participation or prompt vendors to charge more for the risk.

Still, the Professional Services Council sees unresolved issues. One is scope. As written, the draft could apply not only to advanced generative AI systems but also to routine features such as the grammar checker in Microsoft Word. That creates a practical question: Should every ordinary, AI-assisted software function trigger the same contract obligations as a model handling sensitive government data or supporting consequential decisions?

The council is recommending a more targeted definition tied to real-world applications and identifiable risks. It is also questioning flow-down provisions that would require subcontractors to comply with parts of the clause. A prime contractor generally has a direct contractual relationship with its first-tier subcontractors, but visibility and control become harder farther down the supply chain.

Data rights remain another sticking point. Contractors and agencies need clarity about who may use supplied data, analyze it, incorporate it into model improvements, and distribute resulting information. Loose language can create commercial exposure for vendors while leaving agencies uncertain about their ability to inspect and reuse what they paid for.

Speed is the larger backdrop. Gartner-based reporting from CIO found an average government buying cycle of 22 months, compared with 6 to 7 months for many other technology buyers. It also found that 74% of public-sector respondents considered business-case development time-consuming, while more than 76% identified scope changes requiring additional research and evaluation as a major delay factor.

That lag can be especially damaging in AI and cybersecurity, where products, threats, and technical assumptions may change before a contract is signed. Federal News Network has cited federal acquisition processes taking 600 days from solicitation to signed contract. Earlier defense acquisition research from RAND identified technical risk, immature technology, and funding instability as common causes of schedule delays.

Security threats do not follow procurement calendars. The council president pointed to cyberattacks affecting community water systems in at least seven states, including Minnesota and Michigan. Internet-exposed programmable logic controllers used to monitor and control water infrastructure have been a known concern. CISA, the FBI, and the Environmental Protection Agency have issued warnings about the targeting of exposed industrial controllers.

Water infrastructure is also fragmented across municipalities, counties, other public entities, and private operators. Common standards and coordinated purchasing can therefore be difficult to establish. The prospect of another continuing resolution at the start of the upcoming fiscal year, alongside cuts in critical areas at EPA and CISA, adds more uncertainty.

The council president argued that restoring Other Transaction Authority to the Department of Homeland Security could help the agency move emerging technologies into development and prototyping more quickly. Meanwhile, the FAR overhaul has put contractors in what the council described as a “messy middle,” with legacy clauses, class deviations issued last year, and new clauses entering contracts. GSA’s willingness to revise its AI language is an encouraging signal, but the real test will be whether the final rules reduce friction without weakening accountability, competition, or taxpayer protections.