Study Finds US AI Rules Are Thinnest for Banks and Hospitals

A team of eight researchers did something simple but revealing: they gathered 684 U.S. federal documents that touch on artificial intelligence — strategies, guidance, reports, policy papers — and asked two questions about each one. How often does it mention a given industry or risk? And how seriously does it actually discuss it? They scored every document on 14 sectors (banking, healthcare, national security, and so on) and 24 different AI risks. Then they compared those scores to how 272 experts ranked the same sectors and risks by how vulnerable they really are.

The mismatch is the story. Washington's AI writing spends the most energy on technical concerns: whether AI systems are reliable, whether they can be hacked, and whether anyone is clearly in charge of them. It spends noticeably less on social and economic fallout, environmental costs, and "multi-agent risks" — the worry that many AI systems working together could pile small mistakes into one big one. On industries, public administration, national security, information, and science get comparatively thorough treatment. Finance and healthcare get less — even though the experts surveyed rated both as highly vulnerable to AI problems.

So what? Banks already use AI to decide who gets a loan, flag fraud, and spot suspicious transactions. Hospitals use it to read scans, help triage patients, and handle billing. If the federal rulebook is thinner here than experts think it should be, you may have fewer guarantees about when an algorithm makes a call about you, whether anyone checks it, or how you'd challenge a decision that goes wrong.

One honest catch: this study maps documents, not outcomes. Low coverage in a paper doesn't automatically mean AI in your hospital is unregulated — plenty of older rules like medical privacy law and FDA oversight still apply. And the researchers aren't claiming AI has already caused harm; they're pointing at gaps where harm could slip through unnoticed. Still, if you want to know where the safety net is thinnest, this is a useful place to look.