December 13, 2023 — Data & Society’s comment on the Office of Management and Budget’s (OMB) draft guidance applauds the important steps it takes toward AI accountability, and encourages OMB to keep those measures in its final guidance.

As we note, the draft guidance rightly identifies many AI systems that have caused harm to the public, and requires federal agencies to institute practices to protect against further such harms. Critically, it starts with the presumption that certain uses of AI impact people’s rights. It also mandates a floor of minimum practices that an agency must meet in order to use an AI system. The required minimum practices for rights-impacting AI systems include sociotechnical approaches to AI accountability like impact assessments and public consultation, which represent the best known standards for accountability across a wide variety of AI systems. Key to the future effectiveness of this guidance, the OMB memo requires that if minimum practices are not met, or if testing required by these practices reveals concerns, agencies cannot use the system in question.

The procedures described by the draft memorandum are both achievable and necessary to protect the public from the demonstrated harms of rights-impacting AI systems. We encourage OMB to preserve these needed protections in its final guidance, and particularly to maintain the requirement that federal agencies may not use systems that fail to meet the identified minimum practices.