Deb Raji joins Stanford HAI, calls enforcement AI's top gap

Deb Raji's biggest worry about AI policy right now is not catastrophe but enforcement. Asked by Stanford HAI to name the field's hardest unsolved problem, she answered: "Enforcement, and I don't think it's close."

Raji, a visiting fellow at Stanford HAI this fall who will join the center as a fellow and assistant professor in computer science in 2027, calls her research area "AI accountability infrastructure." That means auditing practices, incident monitoring databases, and what she describes as "the government capacity to enforce what's on the books." She came up as a collaborator on Joy Buolamwini's audits of facial recognition bias and helped create Model Cards at Google.

Her view of current policy debates: too much weight on extreme scenarios, not enough on the harms already running. "My honest judgment is that the bar is set too high" for what counts as a significant AI harm, she says, pointing to effects that are "often pernicious and cumulative."

The settings she keeps returning to are the ones where there is no exit. Government, education, healthcare: "You can't opt out of them." Agencies, she notes, "especially in certain states, are operating on a shoestring," which leaves them taking vendor narratives at face value. She points to New York City's Local Law 144, which requires bias audits of AI hiring tools, California's SB 53, and the Frontier AI Act now being debated in Congress as the live tests of whether enforcement can keep pace with deployment.

There is a framing objection, too, about how the conversation gets conducted. "When people talk about AI, we tend to dehumanize the actual humans...and then we anthropomorphize the model itself." Her advice to policymakers is to seek independent technical perspectives rather than lean on corporate representatives.