RICHMOND — The artificial intelligence revolution has arrived in Virginia, but state government is "falling short" in preparing to take advantage of the rapidly advancing technology's benefits while guarding against the risks of its misuse, the General Assembly's watchdog agency reported on Tuesday.
The Joint Legislative Audit and Review Commission outlined opportunities that AI offers to improve state and local government operations, including using the technology to reduce errors in administering the Supplemental Nutrition Assistance Program that could cost the state hundreds of millions of dollars under a new federal law that requires states with high error rates for eligibility to pay a larger share of the program's costs.
But the 97-page study also finds that the Virginia Information Technologies Agency, or VITA, is ill-prepared to either take full advantage of the opportunities that AI presents or establish guidelines for its use that are proportionate to its risks.
"VITA is currently falling short in the support, guidance and oversight it is providing state government in the area of artificial intelligence," Staff Director Hal Greer told legislators on the commission, known as JLARC. "It needs to substantially (step) up its game."
Greer, who is retiring early next year after 32 years at JLARC and 13 years as staff director, also warned legislators that the Division of Legislative Services, which the General Assembly relies upon to pass new laws, doesn't have the expertise in technology law that will be necessary to draft legislation that addresses the challenges that AI technology presents.
"Legislating in the era of AI can be extraordinarily complex and challenging," he said, "and it is imperative that the Division of Legislative Services acquire the specialized technology legal expertise needed to provide the General Assembly with the counsel it should have in this area."
The report recognizes that the potential risks of AI technology extend beyond its handling by state government agencies and employees, but they also affect the public, especially young people, many of whom already use AI chatbots. The Massachusetts Institute of Technology has created an AI risk repository that documents more than 1,700 known risks of the technology, while the FBI has logged more than 22,000 AI-related criminal complaints last year, with nearly $22 million in economic losses in Virginia.
Gov. Abigail Spanberger has created the AI Task Force — led by Secretary of Administration Traci Deshazor, Counsel Matt McGuire and Chief Transformation Officer David Wilkinson — "to help safeguard people and communities," said Deshazor, a former Richmond administrator who oversees VITA and other affected state agencies.
"Public safety must always come first when developing technology, and AI is no exception," Deshazor said in a letter to Greer this month in response to the study's findings.
Greer cited encouraging progress at VITA under Chief Information Officer Michael Watson, who became VITA's leader in April after 19 years at the agency.
Watson, who previously had served as deputy CIO and chief information security officer, acknowledged the challenges that AI poses to state government, as well as the urgency in responding to the rapid advances and deployment of the technology.
"We know AI is everything at this moment, and everywhere," he said.
Watson said, "We know there are a lot of opportunities for us to make changes to what we're doing today ... Our job is to make sure the agencies are ready."
The JLARC study found that about one-third of state agencies are using or testing AI for specific functions, while most state employees use general-purpose AI tools. There is little use of "agentic" AI, in which the technology is empowered to operate semi-independently to make decisions with minimal human oversight.
But it also found that AI requirements for the 65 executive branch agencies that VITA serves "lack risk-based criteria, clarity and timely updates, undermining efficient and effective oversight."
"Policy and training gaps on general-purpose AI tools increase the state's risk of sensitive data exposure and other problems," the study summary states.
The technology offers what JLARC called "near-term opportunities" to modernize outdated state information technology systems and help the Virginia Department of Social Services reduce the error rate in eligibility for public food assistance under SNAP.
The sweeping tax cut legislation that President Donald Trump signed last year increases the state's share of the program's administrative costs and, for the first time, requires states to pay a portion of the benefits themselves, depending on the error rate. Virginia would have to pay up to $270 million a year, beginning in the next fiscal year, if it doesn't lower its error rate. JLARC recommends that VITA prioritize use of AI technology to help reduce the error rate.
The study also cited "data readiness problems and poor coordination within VITA" as barriers to state agencies using the technology.
It also focuses on a lack of expertise in technology law at the Division of Legislative Services that "may challenge its ability to provide effective guidance on AI-related legislation."
Last month, the study said the division had no attorney on its staff who specializes in technology law or access to one.
It said the division, which is part of the legislative branch of government, could hire someone with that expertise, partner with a university or retain outside legal counsel to help it guide legislative efforts to strengthen Virginia laws governing artificial intelligence and other technologies.
The study suggests that the assembly may want to include funding in the state budget "to obtain, on an ongoing basis, specialized technology law expertise needed to effectively adviser legislators on AI legislation."
