A new report from the Center for AI and Digital Policy (CAIDP) reveals widespread non-compliance among federal agencies with mandated artificial intelligence safety measures, potentially putting civil rights and public safety at risk.
CAIDP reviewed the published statements of federal agencies to assess the status of AI systems, and determined that no federal agency has published required determinations, about whether their AI systems impact public safety or civil rights, despite approaching deadlines set by the Office of Management and Budget (OMB).
Of the 50 federal agencies surveyed, only 28 have published compliance plans detailing their current and planned uses of AI systems, according to the CAIDP review. No agencies have submitted extension requests for continuing non-compliant AI systems, which were due by October 15, 2024.
"Rights-impacting and safety-impacting AI systems that fail to meet minimum practices are currently being deployed by federal agencies," the report states, noting that "agencies have not yet acted to address the risk of these AI systems."
CAIDP's review found that the March 2024 OMB memorandum established clear obligations for federal agencies. According to CAIDP's findings, agencies were required to disclose current and planned use of AI systems, determine whether AI systems are rights-impacting or safety-impacting, establish minimum practices to protect civil rights and public safety, and decommission the AI systems that fail to meet minimum practices by December 1, 2024.
Key agencies that have yet to report on their AI activities include the Office of Management and Budget, Office of Science and Technology Policy, and the Consumer Financial Protection Bureau.
The report gains additional significance in light of the United States' recent signing of the Framework Convention on Artificial Intelligence and Human Rights, Democracy, and the Rule of Law in September 2024, which commits the government to ensuring AI technologies promote respect for human rights and democratic values.
"The failure of many federal agencies to publish compliance reports indicating current and planned use of AI systems is concerning," the report concludes. "More concerning is the failure of federal agencies to take steps to decommission AI systems that are rights-impacting or safety-impacting that fail to meet minimum practices."