California Bans AI That Reads Employees’ Moods (AB 1883) — Five Takeaways from Our 2027 New Laws Masterclass

In my final scorecard on the Governor’s signing decisions, AB 1883 got two sentences and one piece of homework: ask your camera-analytics and call-monitoring vendors whether any feature scores “sentiment” or “engagement,” and turn it off for employees. That is still the homework. But at yesterday’s Zaller Law Masterclass on the new laws for 2027, my partner Joshua Dale walked through AB 1883 in more detail, and the questions that came in through the chat showed where employers are actually going to run into this law. It is not where most people expect.

Quick background. On September 30, Governor Newsom signed AB 1883, which adds new provisions to the Labor Code beginning January 1, 2027. Employers may not use a workplace surveillance tool that uses artificial intelligence on employees to (1) recognize, or make inferences or predictions about, an employee’s emotional state, or (2) collect neural data. As Joshua put it, AB 1883 is a vestige of SB 7, the broader “No Robo Bosses” bill the Governor vetoed last year. Most of SB 7 came back this year as the narrower SB 947; the piece aimed at emotional surveillance came back as its own bill.

Here are five takeaways for California employers.

1. The line is mood, not movement.

The most useful framing from the masterclass was Joshua’s: think of AB 1883 as regulating subjective emotional states. The law does not ban video monitoring, and it does not ban AI. Many businesses have to run cameras, and they can keep doing so. Employers can still use AI to track where employees are, whether a break was taken, or what task someone is performing. What employers cannot do is use an AI tool to conclude that an employee looks tired, angry, stressed, frustrated, or disengaged.

Two features of the statute make that line broader than it first sounds:

• Inferences and predictions count. A tool does not need to claim it “reads” emotions. A “burnout risk” flag, a “frustration” score on a recorded call, or a prediction that an employee is disengaged can all qualify.
• “Emotional state” is not defined. Stress, frustration, enthusiasm, and “engagement” are obvious candidates. Fatigue and attentiveness are harder calls (see point 3).

It also helps to know what the law does not reach. It applies only to tools that use AI. A supervisor who watches a camera feed and decides an employee seems upset is not covered. Software that makes the same call is. And the ban turns on how the employer uses the tool, not on what the tool is capable of. An earlier version of the bill reached tools merely “capable of” these functions; the enacted version does not. That matters a great deal for point 2.

2. The risk comes in through your vendors, not your own AI projects.

Very few restaurant, hospitality, or retail employers are building their own AI platforms. As Joshua noted, the exposure is far more likely to arrive through a system someone else rolls out to you: a CCTV upgrade with new “AI enhancements,” a call-analysis tool, a franchisor-required platform, or a national vendor release that turns on a new feature across every customer at once. A system that was giving you objective data last quarter may start scoring tone or mood this quarter, and nobody at the store level will notice.

Samantha Gallagher of Harri, who joined us for the masterclass, made the same point from the vendor side: know which laws apply to the technology you use, not just to your own employment practices, and follow up with your vendors to make sure they understand what is expected of them if there is an audit or a lawsuit.

Because the law focuses on use, the practical fix is usually straightforward:

• Ask in writing. Does the product recognize, infer, or predict anyone’s emotional state from video, voice, or text? Does it collect data measuring nervous system activity? Can those features be turned off for California employees?
• Turn it off and document it. A dated record of which features were disabled, when, and by whom is one of the most valuable compliance documents you can create for this law.
• Get notice of changes. Build a contract term requiring the vendor to notify you before enabling new analytics features of this kind.
• Re-check after every update.Turning a feature off once is not the end of it. Software updates and new releases can add new analytics or reset settings to the vendor’s defaults. Assign someone to review release notes for your camera, call-monitoring, and HR platforms and confirm that emotion-related features are still off after each update.

3. Driver and safety monitoring: watching the eyes is fine; predicting the feeling is not.

The most common question in the chat was about driver safety systems, and it is a good one for any employer with delivery, catering, transportation, or field service drivers. AB 1883 says it does not prohibit an employer from using a workplace surveillance tool “to ensure safety,” and Joshua’s guidance on how that plays out was practical:

• Physical cues are fine. A system that detects whether a driver’s eyes are closing, whether the driver is looking at the road, whether they are on their phone, or whether they are nodding off is monitoring physical aspects of driving, not mood.
• Timers are fine. A system that says “you have been driving for six hours, take a break” is not making any inference about the driver’s emotional state.
• Algorithmic predictions of how someone feels are the risk. If a system starts predicting that a driver is getting tired or stressed from a combination of algorithmic factors, rather than from observable physical cues, it is moving into territory AB 1883 may prohibit.

Employers should know that the safety provision is one of the open questions in the statute. The text protects tools used “to ensure safety, or otherwise if it does not meet the specifications” of the ban, and early commentators have split on whether a safety tool is protected even when it does infer an emotional state or collect neural data. Until a court weighs in, the conservative approach is the one Joshua described: keep the safety tools that monitor physical behavior, watch for vehicle-manufacturer and vendor updates that add “stress” or “mood” detection, and turn those features off. For any tool you rely on as a safety tool, document the specific hazard it addresses, and keep the data out of performance reviews and discipline. A safety rationale is hardest to defend once the data starts being used to manage performance.

4. Wearables: the “neural data” label matters less than what the device concludes.

The second prong of the law bars using an AI-enabled tool to collect neural data, which the statute defines as information generated by measuring the activity of an employee’s central or peripheral nervous system that is not inferred from nonneural information. The clear examples are fatigue-monitoring EEG headbands and “focus-tracking” earbuds. For those devices, collection alone is a violation, whether or not the data is ever used.

The harder cases are ordinary wearables. Joshua held up his own smartwatch during the masterclass, which reports a “stress” score. Technically, a stress reading derived from heart rate is probably not neural data under the statutory definition, because it is inferred from nonneural information. But that distinction does not help the employer much. If the device uses AI to infer stress, it falls under the first prong — an inference about the employee’s emotional state — so the practical answer is the same: an employer should not be collecting or using that score. The bottom line for wearables is to look at what the device concludes about the employee, not just at what it measures.

Wearables also connect to another bill signed this year. AB 1331 bars workplace surveillance in bathrooms (which was already unlawful in California, but now it is explicit) and gives employees the right to leave wearable surveillance devices outside the restroom. As I mentioned during the masterclass, wearables also raise a question employers have not fully thought through: if an employee is being tracked during a meal or rest break, are they really relieved of all duty? Any wearable program should address what the device records during breaks and how employees can turn it off.

5. Call and sales analytics are the gray area.

I have not seen many clients using emotion-detection tools directly. Where I do see it come up is in sales and customer service environments that record calls or video meetings, then grade the representative. Many of these platforms rate whether the customer seemed engaged and coach the representative on how to approach the conversation differently. That is a gray area. A tool that analyzes the guest’s sentiment is a different question than one that scores the employee’s tone, but many platforms do both, and the employee-facing piece is what triggers AB 1883. Joshua’s concern was the same one from point 2: a vendor adds tone or mood analysis to a call-scoring system that used to rely on objective metrics, and you are out of compliance without changing anything yourself.

On enforcement, the Labor Commissioner may investigate, issue citations, and file a civil action, and public prosecutors may also enforce the law. Remedies include injunctive relief, punitive damages, attorney’s fees and costs, and a civil penalty of up to $500 per violation. The statute does not define what counts as a single violation. Earlier drafts included a private right of action, and the final bill removed it, but as Joshua noted, the civil penalty may well make its way into PAGA notices.

That is also why the documentation matters. The tool inventory, the vendor responses, and the record of disabled features are the kind of compliance work that supports a “reasonable steps” position under the 2024 PAGA reforms if a notice ever arrives.

Practical steps before January 1, 2027:

• Inventory features, not just products. Review camera analytics, CCTV, call-monitoring, contact center, video meeting, HR analytics, and wellbeing platforms for anything labeled sentiment, tone, emotion, engagement, mood, stress, wellbeing, or burnout. Separately list any wearables or fatigue-monitoring devices.
• Send the vendor questions in writing and keep the answers.
• Disable and document any feature that infers how an employee feels.
• Review driver and safety systems for “smart” features beyond physical-cue and timer-based monitoring.
• Update monitoring policies and train managers that AI-generated conclusions about how an employee feels are off limits, whether they come from a camera, a microphone, or a chat log.
• Coordinate with the other technology laws. The same vendor review should cover the No Robo Bosses Act (SB 947), which requires human corroboration and notice when an automated decision system drives discipline or termination beginning July 1, 2027, and SB 951, which changes Cal-WARN notices for AI-driven layoffs.

The bottom line: AB 1883 does not stop employers from monitoring the workplace, running cameras, or using AI. It stops employers from using AI to decide how employees feel. For most employers, compliance will not mean changing what they monitor. It will mean finding the sentiment, mood, and “engagement” features their vendors have quietly added, turning them off for California employees, and keeping a record that they did — before January 1, and before a vendor’s next update turns one back on.

Join us in person on November 17. We will be covering AB 1883 and the other new laws for 2027 in more depth at our in-person Employers Summit at the Proud Bird in Los Angeles, along with panels from business operators on the state of employment and the economy in California heading into 2027 – learn more about the Employer’s Summit and register at our website here.