California Just Redrew the Rules for AI at Work

Newsom just signed the country's first law creating state-certified AI auditors. Two more bills are sitting on his desk right now. One would ban AI from reading your employees' emotions. The other says a computer can't be the one that fires

Newsom just signed the country's first law creating state-certified AI auditors.

Two more bills are sitting on his desk right now. One would ban AI from reading your employees' emotions. The other says a computer can't be the one that fires them.

If AI touches hiring, scheduling, performance scoring, or discipline anywhere in your business, California's new AI employee monitoring rules are worth five minutes of your attention. Even if you've never set foot in Sacramento. California writes labor rules the rest of the country tends to copy a year or two later, and this batch has real teeth.

Newsom Signed the Country's First AI Audit Law

On September 9, 2026, Governor Gavin Newsom signed SB 813 and AB 1405, creating the first US framework for state-certified AI auditors.

SB 813, from Senator Jerry McNerney, sets up the California Artificial Intelligence Standards and Safety Commission. Its job is recognizing "Independent Verification Organizations," expert panels qualified to assess whether an AI system meets state safety, reliability, and security requirements. AB 1405, from Assemblymember Rebecca Bauer-Kahan, builds the registry those auditors have to join, with standards for independence and transparency.

Together they reach "high-risk automated decision systems," meaning software that scores, classifies, or recommends outcomes for real people. That's a broad label, and early coverage of the law names hiring tools and insurance pricing as examples that fall inside it. If you bought an off-the-shelf AI tool to help sort resumes, you're the kind of user this framework was built around.

Nobody has to register as a certified auditor tomorrow. The Government Operations Agency has until January 1, 2028, to stand up the certification process, and the requirement that a covered audit actually be performed by a registered auditor doesn't bite until January 1, 2029. This is infrastructure, not an immediate mandate. But it tells you exactly where the state is heading, and it's heading toward "prove it," not "trust us."

We cannot expect industry to simply grade its own homework. Third-party auditors are essential to ensuring AI is safe for our communities and critical infrastructure.

— Assemblymember Rebecca Bauer-Kahan, co-author of AB 1405

Two More Bills Are Waiting for a Signature

The audit framework wasn't the only AI employment legislation to clear the legislature this session. Two bills that hit day-to-day HR decisions directly are still pending.

AB 1883 was enrolled and sent to the governor on September 10. SB 947, the "No Robo Bosses Act," landed on his desk a day earlier, on September 9. Bills presented to the governor after September 1 get a firm deadline: sign or veto by September 30, or the bill becomes law without a signature. Both bills are on that clock.

Neither showed up out of nowhere. Call-center software that scores an agent's "sentiment" in real time is now a normal line item, not a novelty. Wearables that claim to flag burnout before it happens got pitched to HR departments as a wellness perk. Scheduling and performance tools that auto-generate a "coaching recommendation" from a productivity score are baked into more shift-work platforms every quarter. Lawmakers wrote these two bills in direct response to that spread, not in the abstract.

The Emotion-Reading Ban Covers More Than Wearables

AB 1883, authored by Assemblymember Isaac Bryan, would stop employers from using AI-powered workplace surveillance tools to recognize, infer, or predict an employee's emotional state, or to collect "neural data." The bill defines neural data as information generated by measuring the activity of an employee's central or peripheral nervous system, not the kind of thing you'd stumble into by accident.

It's narrower than a blanket surveillance ban. Tools used strictly for safety stay legal, and the bill carves out work tied to aircraft development and national security. But the target is specific: call-center software that scores an agent's tone for "frustration," wellness apps that read stress from a wearable, anything that turns an employee's body into a data stream about their mood.

If it's signed, enforcement runs through the Labor Commissioner, public prosecutors, and employees themselves, who get a private right of action in court. The penalty is $500 per violation, on top of injunctive relief and attorney's fees if an employee sues and wins. That's not a rounding error if a monitoring tool touches every shift, every day.

No Robo Bosses Means a Human Signs Off

SB 947 is a second attempt. Its predecessor, SB 7, passed the legislature in 2025 and Newsom vetoed it, calling it overbroad and duplicative of existing labor protections. Senator McNerney came back with a narrower version. It passed the Senate 29-9 and the Assembly 55-17, a wide enough margin that a second veto would be a real fight, not a formality.

The core rule: an employer can't rely solely on an automated decision system to discipline or fire someone. A human has to review the case and back it with corroborating evidence, not just rubber-stamp what the software flagged. The bill also blocks using an automated system to infer protected status like race, religion, or gender, or to retaliate against workers for legally protected activity, and it requires timely, plain-language notice when a decision leaned on one.

If Newsom signs it, SB 947 becomes operative July 1, 2027. That's later than AB 1883's expected January 1, 2027 start, giving employers more runway, but the direction is the same: automation can assist a people decision, it can't own it.

None of This Requires You to Rip Out Your Tools

The practical answer isn't "stop using AI in HR." It's "stop letting AI make the final call without a paper trail." That's a smaller lift than it sounds, and it's the same discipline we'd expect any regulated automation to have, the same logic behind why AI candidate screening became high-risk under the EU AI Act this year: the tool can do the work, but a human has to own the decision it feeds.

Concretely, that means three things. A documented human-review step before any AI-influenced discipline or termination goes through, not a checkbox nobody looks at. Logging of what the AI flagged and why, so there's a real record if a Labor Commissioner or a plaintiff's attorney asks. And an honest audit of any tool that scores sentiment, mood, engagement, or "burnout risk" from employee data, because that's precisely the category AB 1883 is aimed at.

Run that audit even if you think you're clean. A lot of software ships with a "coaching insight" or "agent wellbeing score" feature turned on by default, one nobody on your team asked for or remembers enabling. If it's inferring how an employee feels rather than measuring what they did (calls handled, tickets closed, units shipped) it's the kind of feature this legislation exists to catch. The fix is usually a settings change, not a new vendor contract. But you have to go looking for it, because it rarely announces itself in a demo.

The applicant-tracking side of this deserves the same look. If your hiring pipeline auto-rejects resumes below a score threshold with no human glancing at the borderline cases, that's the same "solely automated decision" pattern SB 947 targets, just applied earlier in the employment relationship instead of at the end of it.

We build and run automation with exactly that gate wired in: the AI does the sorting and flagging, a person makes the call, and every step leaves a record. It's not a compliance department. It's a workflow that happens to hold up if someone ever asks you to explain it.

California Sets the Pattern the Rest of the Country Follows

This isn't happening in isolation. Fourteen states passed chatbot disclosure laws this year alone, and we've tracked which ones actually reach a business phone line. The EU AI Act already treats hiring and employment-decision AI as high-risk, with its own documentation and human-oversight requirements, covered in our EU AI Act guide for small business. California's employment bills fit the same arc: start with hiring, move to monitoring, move to firing.

You don't need a California address for that pattern to matter. Employment law in the US tends to spread state by state, and California is rarely the last state to pass a given rule, usually the first. Businesses that wire in the human-review gate now aren't reacting to a California mandate. They're building the habit before it's mandatory anywhere else.