Dark console-style card reading Your boss's AI can't guess how you feel. A guess about a feeling is not a measurement.

Your Boss’s AI Can Still Watch You Work. It Can’t Guess How You Feel

by
Reading Time: 5 minutes

The AI can still watch you work. It just can’t guess how you feel about it. Which of those two lines matters more?

California’s AB 1883 bars employers from using AI-driven workplace surveillance tools to infer an employee’s emotional state, or to collect “neural data” from their nervous system. Governor Newsom signed it on 30 September 2026 as Chapter 853, and law firms report it takes effect on 1 January 2027.

That is the whole headline. The interesting part is what the law leaves alone.

🧾 The Two Bans, in Plain English

The new Labor Code section 1581(a) says an employer shall not use a workplace surveillance tool that uses artificial intelligence on employees for either of two purposes. (Chaptered text, California Legislature)

  1. Recognizing, or making inferences or predictions about, an employee’s emotional state. Note the verbs. A tool does not have to claim certainty. A prediction about your mood counts.
  2. Collecting neural data. The law defines this as information generated by measuring activity of an employee’s central or peripheral nervous system, and not inferred from nonneural information.

The second ban is about measuring the body. The first is about guessing the mind. Both are about turning a person into a signal.

Here is the detail that is easy to miss. The law says “uses artificial intelligence,” and it defines AI broadly: a machine-based system that infers, from the input it receives, how to generate outputs. So the question for a buyer is not “is this a chatbot?” It is “does this system produce an output about an employee’s feelings, or collect readings from their nervous system?” If the answer is yes, the label on the box matters much less than the output.

🔍 What Counts as a “Workplace Surveillance Tool”

The definition is wide. It covers any system, application, instrument or device that collects employee data, activities, communications, actions, biometrics or behaviours by means other than direct observation by a person. The statute lists video and audio surveillance, continuous incremental time-tracking, geolocation, and electromagnetic or photoelectronic tracking as examples. (Chaptered text)

“Employer” is wide too: private companies, state and local government, universities, labor contractors, and client employers who use a labor contractor.

📌 Real-world case: The statute names no product and cites no example, so I will not say any named tool is now illegal. But the category is not hypothetical. NiCE, a contact-centre software vendor, says on its own site that sentiment analysis evaluates both customer and agent language, and that its models learn from word choice, tone markers and “acoustic cues in voice interactions.” (NiCE, vendor page) That page describes the product; it does not say whether any particular feature falls under AB 1883. That is for each buyer, and their lawyer, to work out.

✅ What Stays Legal

Read section 1581 carefully and you find a narrow law.

The ban is on AI used for those two purposes. Video, audio, time-tracking and geolocation tools are named in the definition, but AB 1883 does not ban them. Section 1581(b) says the law does not prohibit using a workplace surveillance tool “to ensure safety, or otherwise if it does not meet the specifications in subdivision (a).” (Chaptered text)

So the AI can still count your keystrokes, log your location and time your breaks, as far as this law goes. Other laws may say something about that, and a separate bill, AB 1331, deals with restroom surveillance. This one does not.

That is why the headline of this post is split in two. California did not decide that watching workers is wrong. It drew a line at one kind of watching: the kind that claims to know what is going on inside your head or your nervous system. Everything on the outside of that line, what you did, where you were, how long it took, is still visible to the employer.

It is a useful reminder for anyone who read “California bans AI surveillance” in a headline. It does not. It bans two specific uses of it.

⚖️ The Line That Will Be Argued

Look again at that quoted sentence: “to ensure safety, or otherwise if it does not meet the specifications in subdivision (a).”

Does the safety clause protect a tool that infers emotion, as long as the purpose is safety? A fatigue-detecting headset is the example people reach for.

Law firms read it differently. Fisher Phillips calls the question unsettled and sets out a narrow reading and a broad one. Mintz reads it narrowly, as allowing safety tools only where they avoid the banned conduct, and sees no wellness or stress-monitoring exception. (Fisher Phillips, Mintz)

I am not going to pick a side. Seedhi baat: the text supports both readings, and no court has decided yet. If a vendor tells you the safety clause covers their product, ask them to put that in writing.

🏛️ Who Enforces It, and What It Costs

Enforcement sits with the Labor Commissioner and public prosecutors. They can investigate, issue citations and bring civil actions, and the penalty is up to $500 per violation. The statute does not define “violation,” so it is unclear whether one tool, one employee or one day is the unit. I will not guess a total. (Chaptered text)

Law firms also say the law gives employees no private right of action. Ogletree adds that employers could still face exposure under California’s Private Attorneys General Act. (Ogletree)

There is a narrow exemption for developing aircraft for the national airspace and for national security, military, space or defense work, and only where the use is reasonably necessary to comply with federal law or a binding federal contract.

🪙 Token Truth

A guess about a feeling is not a measurement.

An emotion score is an inference sold as data. The tool never sees your mood. It sees a voice pitch, a word choice, a heart-rate trace, and a model turns that into a label. It is the same trap as counting tokens and calling it productivity: an easy number standing in for something that is hard to measure. (If you want that thread, start with Episode 1 of The Token Diaries.)

California’s law does not rule on whether such scores are accurate. It simply says an employer may not use AI to produce them.

It also connects to a worry I raised in Worker Optimization: when a number becomes the target, people start performing for the number. A tool that scores your mood invites you to perform a mood. And if you are wondering about the data you hand over in a different way, Your Prompts, Their Data covers what prompt logs reveal.

🛒 If You Buy or Build Monitoring Tools

Five questions to ask before 1 January:

  1. Does it score mood, stress, sentiment, engagement or fatigue on employees?
  2. Does it use wearables or sensors that measure the body?
  3. Is “wellness” part of the sales pitch?
  4. Which safety purpose would you actually defend, and in what words?
  5. What does the vendor say in writing about how it fits AB 1883?

The clean answer to the first two is the one that keeps you well away from the line.

Two practical notes. First, ask for the answer in terms of outputs, not features: what does the dashboard actually show about each person? A “wellbeing score” and a “stress indicator” are exactly the kind of output the law is aimed at. Second, check the whole supply chain. The definition of employer includes a client that contracts with a labor contractor, so “our staffing partner runs that tool” is not an obvious way out.

🙋 If You Are an Employee

You can ask HR three things: which monitoring tools are in use, whether any of them estimate emotion or collect nervous-system data, and what the safety purpose is for any that do. You do not need to be hostile. These are ordinary questions about your own data.

None of this is legal advice. It is my reading of the text and of what law firms have published, and I have not found guidance from the Labor Commissioner yet.

If a tool could tell your manager you were stressed, would you want it to be wrong, or right?

📚 References

What’s your Reaction?
+1
0
+1
0
+1
0
+1
0
+1
0
+1
0
+1
0