Menu

Navigating the GSA’s Proposed AI Procurement Clause: Unpacking the Operational Impact on Contractors

Posted on March 17, 2026
The General Services Administration (GSA) has introduced a proposed clause, GSAR 552.239-7001, titled “Basic Safeguarding of Artificial Intelligence Systems,” aiming to regulate the procurement and use of AI systems within federal contracts. This initiative aligns with the Office of Management and Budget’s (OMB) Memorandum M-25-22, emphasizing the acquisition of “American AI Systems”—those developed and produced in the United States.
The proposed clause represents one of the most consequential shifts in federal AI procurement to date, with far-reaching implications for how contractors develop, deploy, and manage AI systems under GSA contracts. With GSA signaling its intent to formally publish Refresh 31—which would incorporate this clause directly into GSA contracts—as early as March or April 2026, the window for industry to shape this policy is critically narrow. Contractors and Service Providers have until March 20, 2026, to submit comments via to maspmo@gsa.gov or through GSA’s Advanced Notice for MAS Refresh 31 blogpost.
Many of the key changes stem from expanded definitions. This blog dissects the core elements of the proposed clause and sheds light on the practical consequences contractors must prepare to face.

1. Expanded Intellectual Property Rights: A Double-Edged Sword

While the proposed clause allows contractors and Service Providers to retain ownership of their underlying AI systems and models, the clause requires contractors to grant the government an irrevocable license to use AI systems for any lawful government purpose. More alarmingly, it awards the government ownership over “Custom Developments” — broadly defined to include any modifications, customizations, configurations, or enhancements to the AI system, expressly including modifications resulting from AI training, as well as related workflows, work product, and deliverables.
Government data is defined to include both “data inputs” and “data outputs.”

What this means operationally:

  • Contractors risk losing control over innovations refined or improved during government contracts, potentially jeopardizing the commercial value and competitive edge of their AI models.
  • Vendors might limit offerings to government contracts or create costly, less competitive “government-only” versions, fragmenting their product lines and inflating costs.
  • This raises serious concerns about disincentivizing industry participation and stifling innovation. Requiring an irrevocable license would be inconsistent with commercial item procurement principles under FAR Part 12.

2. Heightened Responsibility for Service Providers

The clause extends compliance and reporting obligations beyond prime contractors to include “service providers.” Under GSA’s proposed definition, a Service Provider is not a party to the contract but either “directly or indirectly providers, operates, or licenses an AI system used in performance of the contract.” At a minimum, this extends requirements to entities that have no privity of contract with the U.S. government.
Operational challenges here include:
  • Contractors must enforce stringent government requirements through complex subcontractor chains where control and visibility are limited or non-existent. Typically, prime contractors do not have visibility or control over the training data, model architecture, or development processes of upstream AI providers. Requiring independent verification of such information would be impractical.
  • Shifting the enforcement burden onto primes will create enforcement challenges that could delay contract performance or distort vendor relationships.

3. Safety Guardrails and Use Restrictions

The government’s right to unrestricted use—free from any vendor “discretionary policies”—could override embedded safety and ethical guardrails designed by AI developers. The proposed clause expressly states that “the AI system must not refuse to produce data or outputs or conduct analyses based on the Contractor’s or Service Provider’s discretionary policies.”
Why this matters:
  • The definition of “discretionary policies” remains unclear.
  • Safety mechanisms built during model training are essential for preventing harmful or biased AI outputs. GSA fails to recognize that this makes it difficult to cleanly separate or disable safety mechanisms and policies without materially altering how a model functions.
  • Removing or softening these protections risks unsafe AI deployments, creating operational hazards for both the government and contractors.
  • Negotiated safety provisions should be reserved for only the highest-risk applications to preserve these vital protections.

4. Data Management: Segregation, Deletion, and Compliance

The clause mandates strict segregation of government data from commercial data, prohibits using government data for model training, and requires data deletion with certification post-contract. Moreover, vendors would be prohibited from “targeting government or non-government entities or informing a contractor’s or vendor’s advertising, marketing, sales, monetization, strategy, operations, or other business decisions.”
Practical implications include:
  • The no-advertising clause represents a notable departure from standard commercial data practices, explicitly prohibiting contractors and service providers from leveraging government data — including AI inputs and outputs — to inform advertising, marketing, monetization, or broader business strategy. This restriction signals the government’s intent to firewall federal AI engagements from commercial exploitation, but it also imposes significant operational and architectural burdens on vendors whose AI systems are built around data-driven business models.
  • Significant investments in data infrastructure and governance are necessary to meet these requirements.
  • Contractors face increased operational complexity and may need to redesign AI architecture that depend on continuous learning.

5. Supremacy Over Commercial Licensing Terms

Traditionally, government contracts respect commercial license agreements to the extent possible. This clause flips that convention: government terms will override any conflicting commercial agreements.
The operational fallout can be profound:
  • Contractors face legal uncertainty and elevated compliance risk navigating this reversal.
  • It limits flexibility in managing risk, liability, and security, undermining vendor confidence.

Strategic Considerations for Contractors

The GSA’s proposed AI clause reflects a strong governmental objective to control and secure AI usage—but it also poses potential barriers to innovation and market participation. To navigate this evolving landscape, contractors should:
  • Conduct thorough AI usage audits across supply chains.
  • Re-negotiate subcontract and service provider agreements to clarify compliance.
  • Invest in robust data governance to segregate, track, and securely delete government data.
  • Actively engage in GSA’s comment period, providing detailed, technically grounded feedback.
By confronting these challenges head-on with clear-eyed realism and strategic agility, contractors can protect their innovations, maintain compliance, and sustain access to federal AI opportunities.
This proposed clause is more than a policy update—it’s a call to align operational capabilities with regulatory realities. This is likely only the beginning. This AI clause may start at the GSA but could become the basis for other agencies to develop AI procurement policies and eventually land in the FAR and DFARs. The outcome will shape government AI procurement and the defense of innovation for years to come.