Draft Federal AI Contract Rules Could Shift Intellectual Property Rights and Add Reporting Burdens
The U.S. General Services Administration (GSA), the agency that manages federal government contracts and procurement, has drafted a new clause for contracts dealing with artificial intelligence (AI) technologies that could limit businesses’ intellectual property rights and impose new, burdensome requirements on vendors.
The “Basic Safeguarding of Artificial Intelligence Systems” clause attempts to establish safeguards and standards for the use of AI in federal procurement and contracts. Elements of the clause govern how data should be managed, who can access government data, how data breaches are reported, testing methodologies, and how AI systems are updated. However, the document also includes several requirements that business advocacy groups, such as the U.S. Chamber of Commerce, have warned could limit competition, increase costs to the government, and discourage AI providers from participating in the federal marketplace.
A critical concern for businesses is language in the clause that appears to shift ownership of modifications developed for AI systems from the companies that develop them to the federal government. As defined by the clause, a custom development is any design of or modifications, customizations, configurations, or enhancements to AI systems or associated implementations or workflows. This would also cover any related work product or deliverables created for the government under the contract. For AI contracts, the clause states that the government retains full ownership of not only government data, but also all custom developments. According to the U.S. Chamber, this draft language could be interpreted to grant the government ownership over fine-tuned models, configurations, and enhancements derived from a contractor’s proprietary technology and pre-existing intellectual property.
Complicating the IP issue, the clause uses overly broad definitions for “AI systems” and “AI service” providers. As drafted, the clause would encompass not only technologies that are commonly understood to be AI, such as large language models and generative AI, but also back-end infrastructure, security monitoring tools, and machine learning algorithms embedded in standard commercial software. In a response to the GSA, the U.S. Chamber said the expansiveness of the scope “introduces significant compliance uncertainty and risks unintentionally including a wide array of commercial offerings that fall outside the intended focus of this regulation.”
The clause’s reporting requirements are another area of concern. If adopted, contractors would be required to disclose all AI systems used in fulfilling the contract, including possible modifications. This could force businesses to report and track AI within common off-the-shelf business tools, including software that has incidental or embedded AI features that are difficult to identify.
Tracking AI systems could be especially important as the clause places sole responsibility for ensuring the compliance of all service providers on the prime contractor. This includes AI systems the contractor neither owns nor operates, which the U.S. Chamber said would significantly expand liability beyond established commercial practices and proven frameworks.
Finally, the clause prohibits the use of foreign AI systems, stating that all AI tools and their components must be developed, manufactured, and controlled by U.S.-based entities. While intended to promote American-made AI systems, the U.S. Chamber pointed out that this definition lacks clarity and could unintentionally restrict the use of open-source technologies, globally developed components, and other widely adopted commercial tools such as those used for research and development.
To improve the draft language, business advocacy groups are encouraging the GSA to more narrowly define terms such as “AI systems,” better align the language with existing commercial practices and frameworks, and clarify that contractors retain all rights to their pre-existing intellectual property.