GAO Flags Privacy Gaps in Federal AI Guidance
The Government Accountability Office called on the Office of Management and Budget to expand its artificial intelligence guidance, saying the current rules fail to address eight of ten major privacy risks that federal agencies face when deploying AI systems.
In a report released Thursday, GAO convened a panel of experts who identified privacy challenges across government AI deployments. OMB's existing guidance fully addressed only two of these challenges.
What Privacy Issues Did GAO Identify?
The gaps center on practical implementation. OMB guidance does not clearly explain how agencies should evaluate and audit AI systems that handle sensitive information or how to isolate sensitive data before using it to train models.
GAO also found that OMB guidance lacks standardized performance metrics and incentives to push organizations toward stronger privacy protections. The rules do not adequately address transparency around user consent or how to incorporate AI-specific considerations into privacy impact assessments.
What OMB Should Do Next
GAO recommended that OMB specify known privacy risks agencies must consider when building AI policies. The watchdog also called for additional guidance on best practices for evaluating and auditing models, separating sensitive data, and establishing performance metrics.
OMB should also improve transparency requirements and integrate AI-specific language into existing privacy impact assessment processes, GAO said.
Recent OMB Requirements
The GAO report follows OMB's release of new requirements earlier this year. In a memorandum, OMB directed agencies to ensure that AI systems they procure meet "unbiased AI" principles.
The guidance requires agencies to collect specific information from vendors about large language models. Vendors must disclose acceptable use policies, system and data cards, end-user resources, and mechanisms for user feedback.
OMB cautioned agencies against requesting sensitive technical details such as model weights, citing the need to balance transparency with intellectual property protections.
Federal employees overseeing AI procurement or policy should review the AI Learning Path for Policy Makers to understand how these requirements apply to their work. Additional resources on AI for Government can help agencies navigate implementation.
Your membership also unlocks:
