The Government Accountability Office has developed a framework that can help the Department of Veterans Affairs, along with other agencies, ensure responsibility and accountability when designing, developing, deploying and monitoring AI systems, according to Carol Harris, the director of IT and cybersecurity at GAO.
Thought leaders in government healthcare technology will discuss the most pressing topics in healthcare technology and citizen user experience at the Potomac Officers Club’s 2025 Healthcare Summit, which is set to take place on Dec. 4. Sign up now to attend this GovCon event, which will also help facilitate new business opportunities and strengthen existing partnerships.
GAO AI Framework
The framework is organized around the principles of data reliability and quality and AI system governance. It also covers the assessment of AI system performance and the monitoring of AI system reliability over time, Harris said in a prepared testimony for an upcoming hearing with the House Committee on Veterans’ Affairs Subcommittee on Technology Modernization.
Each principle is accompanied by key practices, whose elements, if taken into consideration, “can help VA guide the performance of assessments and audits of agency AI implementation,” Harris goes on to say.
VA Challenges With Generative AI
The GAO official also reports the challenges VA faces in using and managing generative AI—challenges that it shares with many other agencies. These issues include compliance with existing federal guidance and policies, the insufficiency of technical and financial resources, AI workforce hiring and development, data security and the acquisition of the AI tools themselves.
Harris noted that VA is in the process of addressing its difficulties with generative AI, in particular by implementing the new AI requirements listed in the April 2025 memorandum issued by the Office of Management and Budget.
“As the AI policy landscape evolves, agencies—including VA—are developing and updating their own guidance intended to govern their use and management of generative AI,” Harris said.
