# GSA’s draft AI procurement rule has improved but needs further reforms, contractors say > Source: > Published: 2026-07-15 19:12:00+00:00 # GSA’s draft AI procurement rule has improved but needs further reforms, contractors say ## Contractors have until Aug. 3 to give GSA feedback on a proposed overhaul to large language model procurements. The General Services Administration’s proposed large language model-specific acquisition regulations unveiled last month are an improvement over rules introduced in March, contractors and experts told federal regulators during a Tuesday listening session. But those stakeholders also told GSA officials that the draft rule requires further modifications that clarify technical definitions and align with commercial best practices. “Stakeholders asked for clarity, practicality and alignment with commercial norms, and this updated draft shows meaningful movement in all of those directions,” Amy Benson, government affairs vice president for Science Applications International Corporation, said at the listening session. GSA published a [draft “Basic Safeguarding of Artificial Intelligence Systems” rule](https://buy.gsa.gov/interact/system/files/GSA_Federal_Acquisition%20Service%20Proposed%20Government%20AI%20System%20Terms%20and%20Conditions.pdf) in March before releasing a [revised version](https://www.federalregister.gov/documents/2026/06/17/2026-12205/general-services-acquisition-regulation-acquisition-of-information-and-communication-technology) titled the “Basic Safeguarding of Data within Large Language Model Artificial Intelligence Systems (LLMs)” last month. If finalized, the proposed rule would require contractors to follow data protection and intellectual property safeguards when they use large language models to process government data. For example, the draft stipulates that contractors cannot use government data to train or fine-tune large language models or inform business decisions. The proposal applies to GSA’s government-wide contracts such as the Federal Supply Schedule, government-wide acquisition contracts and the OASIS+ professional services vehicle. The latest GSA draft softened language around contractors’ use of foreign AI components, assigned flowdown requirements based on four general roles within the industry — LLM Developers, LLM System Operators, LLM System Integrators and LLM Service Providers — and axed a specific prohibition of “diversity, equity and inclusion” ideology in AI models in favor of more generally prohibiting “ideological dogmas.” Contractors said they welcome many of those changes but want GSA to clarify definitions and reform flowdown requirements. “We would argue that the definition of data, although improved, is still overly broad and must be tailored to align with commercial practices,” Megan Peterson, a senior vice president at the Information Technology Industry Council, said at the listening session. She added that the definition of data outputs — which the draft sets as “all data, information, [personally identifiable information], any improvements, enhancements, corrections, annotations, or other modifications made to Data Inputs, or content generated by the LLM in the performance of this contract” — is unclear and could create confusion as to when information becomes government data that requires additional safeguards. Peterson urged GSA to ensure the definition of data outputs excludes metadata, logs and other types of information. Tim LeMaster, vice president of federal engineering at Lookout, similarly told *Nextgov/FCW* that GSA should clarify what "incidental" large language model functionality exempt from the proposed rules means in practice. Benson said the proposal’s definition of custom development — which refers to “modifications, customizations, configurations, or enhancements” made in service of the government contract — is too broad by continuing to cover reusable integrator-developed intellectual property. Another concern Benson raised is that unbiased AI principles language and testing requirements are unclear. “We just totally nailed the unbiased AI principles,” Jeff Koses, a senior GSA procurement executive, jokingly told listening session attendees. “[I] heard that we need to continue to work on definitions and recognize what falls outside of those definitions… [I] heard quite a bit of inconsistency with various commercial terms, we need to work up to data rights, including metadata,” he added. Contractors have until Aug. 3 to comment on the draft rule. GSA has only received 16 comments as of Wednesday afternoon, but several industry groups told *Nextgov/FCW* they plan to submit comments in the coming weeks. As GSA mulls industry feedback, experts are urging the agency to offer examples of how the large language model proposal would work in practice. “Adding examples for what's in and what's out, I think that that's generally a good practice, especially with new regulation,” David Timm, a partner at Burr & Furman, said in an interview. But regulators should also consider a three-pronged test to determine what data requires additional protections under the proposal, Jessica Tillipman, associate dean for government procurement law for George Washington University law school, said.