California AI Transparency Law: State Agencies Hid Six Systems

Sacramento passed a law to expose risky government AI. State agencies simply said “no” — until they got caught.

As California leans harder on automated systems to decide who gets housing, health care, and cash assistance, millions of residents are asking a simple question: if the state can’t even tell the truth about which computers are making these calls, who’s actually accountable when they get it wrong?

Do you know if a machine decided your last interaction with a state agency? Most Californians don’t, and that’s by design.

This isn’t hypothetical. This week, a CalMatters commentary revealed that California’s flagship AI transparency law produced exactly one result in its first year: a spreadsheet where every single state agency checked “no” next to the question of whether it used high-risk automated decision systems. Then, a year later, the state quietly admitted the truth — at least six such systems were already running, deciding who gets Medi-Cal benefits, In-Home Supportive Services, housing help, and more.

What Was AB 302 Supposed to Do?

Assembly Bill 302, signed three years ago, ordered the California Department of Technology to build a “comprehensive inventory of all high-risk automated decision systems” used anywhere in state government and publish the findings every year. The idea was straightforward: if a machine is going to decide whether a family gets food assistance or a disabled Californian keeps home care hours, the public deserves to know it exists.

Instead, the law’s first report found nothing. Not one high-risk system, anywhere in the entire California state government. Watchdogs who filed public records requests got back a single spreadsheet with one column — “Is ADS used” — and the word “no” typed next to every agency’s name. There was no sign the department had asked a single follow-up question.

Why Did the State Suddenly Change Its Story?

A year later, after continued public pressure, the Department of Technology did something it apparently hadn’t done the first time: it actually interviewed agencies. A handful came forward and admitted they were running high-risk automated systems after all — at least six of them, by the CalMatters authors’ count.

This matters because the law depends entirely on agencies grading their own homework. There’s no independent audit built into AB 302, no state office cross-checking agency claims against reality, and no penalty on the books for an agency that simply says “no” when the answer is yes. That’s not oversight. That’s an honor system applied to the machinery of a $200-billion-a-year government.

What Systems Were Never Reported?

Even the state’s revised, six-system count may understate the problem. The commentary names two automated tools that still haven’t been formally disclosed under AB 302: the Uniformity Assessment System, which has been tied to reductions in In-Home Supportive Services hours for disabled Californians, and the Risk Segmentation, Stratification, and Tier model used by state health officials to flag Medi-Cal recipients as high-risk or “underutilizing” services.

Neither tool made it onto the state’s official inventory. If systems already known to researchers and litigators can slip past a “comprehensive” state review, it raises an obvious question: how many more haven’t surfaced yet?

If a machine can quietly help decide your Medi-Cal risk score and the state never has to admit it exists, who is actually governing California?

Who Is Really Paying for This Policy?

Every Californian who depends on a state benefit is, in effect, a test subject for systems that were never supposed to operate in the dark. The people most exposed are also often the least equipped to challenge an algorithm they don’t know is deciding their case — elderly and disabled residents relying on home care hours, and low-income families whose access to food or medical assistance can turn on a score they’ve never seen.

There’s also a straightforward fiscal-accountability angle here. Building AB 302’s inventory and reporting apparatus cost taxpayer dollars. What the public got in return, for at least one full reporting cycle, was a spreadsheet full of “no.” That’s not limited, efficient government — it’s a compliance exercise that produced compliance paperwork and nothing else.

Is a law that depends entirely on the government policing itself really a transparency law — or just the appearance of one?

Is This Just a California Problem?

No. The commentary’s authors point to New York’s Public Oversight of Surveillance Technology Act as a similar cautionary tale — a law meant to expose the NYPD’s use of surveillance tools that instead let the department describe its own capabilities in terms vague enough to dodge real scrutiny. The pattern repeats nationally: lawmakers pass a disclosure requirement, agencies self-report, and self-report becomes self-protect.

A transparency law with no verification and no penalties isn’t oversight — it’s an honor system for the people deciding your benefits.

What Do Supporters of This Policy Actually Believe?

Backers of AB 302 make a fair point worth engaging directly: transparency laws are often the only politically achievable first step, and a flawed inventory is still better than no inventory at all. They argue that public pressure — like the CalMatters records requests that forced this year’s partial admission — shows the law is working exactly as designed, slowly, through accountability journalism and advocacy rather than built-in enforcement.

That argument has real merit, and it’s true the law forced information into the open that would otherwise have stayed hidden. But it also concedes the core problem: the law only worked because outside researchers did the state’s job for it. A transparency statute that requires nonprofit watchdogs to function as its enforcement arm isn’t a durable accountability mechanism — it’s a stopgap. Real government oversight shouldn’t depend on whether a journalist happens to file the right records request in a given year.

Key Questions This Story Raises

• Who verifies that state agencies are telling the truth about their own AI use — and what happens when they don’t?
• How many additional automated systems, like the two named in this report, remain undisclosed right now?
• Should lawmakers move past disclosure-only laws toward binding limits on which government decisions AI is allowed to make at all?

218 million dollars. That’s roughly what California’s Department of Technology has requested in recent budget cycles for oversight and compliance functions tied to laws like AB 302 [state budget estimate]. The question no one in Sacramento wants to answer: what did that money actually buy Californians the first year?

What Happens If Lawmakers Don’t Act?

Without real verification and real penalties, AB 302 will keep functioning the way it has so far — as a formality agencies clear once a year, regardless of what’s actually running in their systems. The people affected won’t be lawmakers or bureaucrats. They’ll be the residents whose benefits, housing applications, and health coverage are quietly shaped by code no one outside government has reviewed.

Californians didn’t ask for algorithms to help decide who eats, who gets housing, or who gets medical care. They did ask, through their elected representatives, to at least be told when that’s happening. Right now, the state can’t even guarantee that much.

The real question isn’t whether California has AI quietly running its bureaucracy — it’s whether anyone will be held responsible for hiding it in the first place.

What do you think — should California scrap AB 302 and start over with real enforcement, or is disclosure-only regulation still worth saving? Share this article and let us know.

Still have questions about how AI is shaping decisions in your state benefits? Stay informed — subscribe to The Town Hall News for daily accountability coverage. Think your neighbors need to see this? Share the article. Want your voice to count? Contact your state Assembly member and ask whether they’ll support real audit and enforcement provisions for AB 302.