GSA’s Revised AI Clause: Key Changes You Should Know About

Earlier this year, the General Services Administration (“GSA”) released a proposed AI clause that had the potential to fundamentally change how contractors implemented artificial intelligence into future federal contracts. As we discussed in our previous article, the proposed clause introduced sweeping requirements related to data ownership, government oversight, and AI governance that left many contractors questioning whether commercial AI systems could realistically comply.

Following substantial feedback from industry, GSA released a revised version of the clause. With many of the provisions being narrowed or clarified, what’s clear is the government is not backing away from AI regulation. Instead, GSA appears to have refined the clause into something that is more practical, signaling its intent to move forward with this requirement very soon.

Below are some of the most significant changes made to the clause and a breakdown of what these changes could mean for contractors moving forward.

A Narrower Scope

One of the largest criticisms of the original clause was that it appeared to apply to virtually every product or service that incorporated artificial intelligence. Given how quickly AI capabilities are being integrated into everyday commercial software, many wondered whether simply using an AI-enabled application would trigger the clause. The revised clause addresses this concern head on. 

Rather than applying broadly to all AI systems, the clause now applies primarily to contractors who intend to use Large Language Model (“LLM”) to process Government Data. There are some exceptions to LLM use, including LLMs that are embedded inside commercial software or to situations where LLM functionality is merely incidental to the operation. 

The AI clause’s new, narrower scope ultimately provides contractors with a much clearer parameter for compliance. Instead of wondering whether a particular software or product falls within the clause’s reach, contractors can now ask a more straightforward question: Are we using an LLM to process Government Data? If the answer is yes, the clause will likely apply. If not, many contractors will likely fall outside the scope of the clause altogether. 

Ultimately, GSA’s revision provides welcome clarity and signals that the main focus is to regulate generative AI rather than artificial intelligence as a whole.

Subcontractor Compliance Overhaul

The biggest structural change to the AI clause was the introduction of four distinct service provider roles.

While contractors are still responsible for service provider compliance with the AI clause, the revised version breaks down the service provider role into four categories: LLM Developer, LLM System Operator, LLM System Integrator, and LLM Service Provider. For each distinct role, the contractor must also implement distinct flow down provisions specifically curated by GSA and directly applicable to each role. 

By dissecting the service provider role into four parts and developing four distinct flow down provisions for each role, the clause will utltimately force contractors to determine which role(s) fit best with each particular subcontractor. And, once the clause is finalized, designation and assignment of these tailored roles will inevitably be a hot topic in subcontract negotiations moving forward. 

“American AI Systems” Requirement Softened

Unlike the initial draft, GSA’s revised AI clause now merely encourages contractors to “maximize” the use of LLMs developed, managed, and operated by a U.S. company (i.e., use of American-made AI systems is no longer explicitly required). This strong preference for American-made systems, however, does come with some handcuffs.

For starters, if a contractor intends to use a non-American-made AI system, the system cannot be controlled by any foreign government. The clause also prohibits use of any LLM and respective components used to perform core model, data storage or processing, output generation, or security functions if the LLM and/or components are developed, managed, or operated by entities subject to the direction, influence, or control of an adversary foreign government (i.e., China, Cuba, Iran, North Korea, Russia (see 15 CFR 791.4)).

While GSA no longer requires contractors to use AI systems exclusively developed in the U.S., the revised clause continues to instill a preference for U.S. systems and places significant restrictions on foreign-controlled systems. Contractors will need to understand not only where their AI models originate, but also who develops, manages, operates, and controls the underlying technology, as well.

Industry Feedback Produced Meaningful Changes

Perhaps the most encouraging aspect of the revised clause is not any individual provision, but the fact that the GSA appears to have carefully considered the concerns raised during the initial public comment period.

For instance, the clause now allows the parties to place a percentage-based cap on decommissioning costs incurred as a result of a termination for cause. This welcome addition means that the contractor may not be fully liable for all decommissioning costs as they were prior to these revisions. This change was undeniably made as a result of industry feedback.

The revisions also tackled due process issues lurking within the government benchmarking and auditing provisions. Initially, contractors had no way of knowing what benchmarks would be implemented in the government’s compliance assessments and evaluations. Now, the clause requires the government to disclose sufficient information to the contractor prior to taking any adverse actions based on lack of compliance with the clause. While questions remain about how benchmarking standards will ultimately be applied, the revised language is a welcomed change and one that provides contractors with an opportunity to address compliance concerns prior to facing potentially severe contractual consequences like contract termination.

GSA also overhauled its AI bias standards. The new standard prevents contractors from intentionally introducing or embedding partisan or ideological judgments into training data, prompts, AI configuration, or retrieval-augmented generation (“RAG”). This change removes the original (and unnecessary) reference to DEI, making the clause considerably more neutral.

Comments on the revised AI clause are currently due on or before August 3, 2026. If you feel compelled, submit your comments before this deadline!

Final Thoughts

Whether GSA’s AI clause is finalized in its current form or revised again, one thing is clear: AI is quickly evolving from a business efficiency tool into a matter of contract compliance. While GSA has significantly narrowed the scope of its AI clause, contractors should not mistake these revisions as a retreat from AI regulation altogether. Rather, the updated clause reveals the agency’s goal to establish AI requirements that are practical, enforceable, and that can be implemented quickly.

As such, contractors should start evaluating their AI systems now rather than waiting for GSA to issue its final rule. Part of this analysis should include identifying each LLM used during contract performance, reviewing agreements with AI vendors and subcontractors, and, if necessary, evaluating whether existing or alternative AI environments can satisfy the updated requirements. Doing this analysis now will ensure that contractors have a leg up when it comes time to compete under these new requirements.

If you have questions about how this clause may impact future bid opportunities or if you want to know more about how to submit or structure formal comments, please contact us for a free consultation.