NEWSLETTER
The Software at Your Job May Be Scoring Your Mood Without Telling You
November 10, 2026

NEWSLETTER
November 10, 2026

California's AB 1883, signed September 30, 2026, restricts employers from using AI-powered workplace monitoring to infer workers' emotional states from January 1, 2027, with exceptions. An AI inventory built from purchase orders can miss emotion-analysis features that arrive through updates to software the company already owns. The edition asks employers to identify which active features infer a worker's emotional state, who enabled each one and where that decision was recorded, and who can change the setting and how a change would be noticed.
Picture a situation at a California insurance company. The company and the people are invented. The kinds of tools and the law are real.
Carla works the phones at a home insurance company in Sacramento, which employs about six hundred people. She has been there for four years and is good at the job. The people calling her are usually having a bad day. A pipe has burst, or a tree has come through the garage roof, and they want to know what their policy covers.
Her screen has the caller’s policy open in one window. Beside it is a thin colored bar that appeared a couple of years ago. It was never explained to her team. It is nearly always green, and Carla has assumed it has something to do with how busy the phones are.
Then her mother is admitted to the hospital.
Carla comes to work, takes her calls, files the claims, and meets her usual quality targets. At about two in the afternoon, the bar turns amber. Underneath it, a message appears: “Try to sound more upbeat.”
She reads it again. The bar has been judging how she sounds.

On Friday, her supervisor asks if he can speak with her for ten minutes. He has a weekly report. It is something he never talked to her about before; there is a row for every person on the team. Her row has three entries: ‘sad or distressed.’
He is kind about it. He asks whether everything is all right at home.
Sitting there, Carla realizes that her manager has seen a machine’s judgment about how she feels before she has told anyone at work.
For how long, she asks.
Approximately two years.
Who decided to use it?
He does not know. It came with the phone system.
Carla signed the employee handbook. It says calls may be recorded for quality and training, much like the announcement a customer hears before a call. It says nothing about software deciding how she feels.

In December, the company’s lawyer gets an alert from an outside law firm. California has passed AB 1883.
Starting January 1, 2027, with some exceptions, an employer may not use AI-powered workplace monitoring to recognize, infer, or predict an individual’s emotional state. Violations can carry penalties of up to $500 each, in addition to other remedies. The law also restricts AI-powered collection of neural data, which is outside the scope of this article.
She forwards the alert to Raj.
“Do we have anything like this?”
Raj manages the IT and inventory for AI. He checks it. There are three entries - the website chatbot, a coding assistant, and a meeting-summary tool. None is listed as an emotion-analysis tool. The company has never bought a product sold as one.
He tells her they are clear.
The inventory does not contain the feature on Carla’s screen, because Raj built it from approved AI purchases. The feature arrived through an update to the company’s existing phone system.
Six years earlier, the company bought a phone system to route calls. Two years ago, the supplier released an update that made voice analysis available. A call center team lead found the setting, thought it might help with coaching, and turned it on.
The change never went through the company’s review process. It never reached Raj’s inventory.

Three days before Christmas, Raj is in a meeting about call center performance when Carla’s supervisor mentions the weekly mood report.
Raj asks which report he means.
Turning the feature off takes one click. That is the easy part.
Eleven supervisors have permission to change the setting. Raj has switched it off, but he has not yet restricted their access or arranged an alert if somebody switches it back on.
He also has to look beyond the phone system. There is an “engagement” score in the meeting software and a “burnout risk” flag in the HR survey tool. If a phone-system update could introduce this capability, those features need examining too.
Raj does not yet know which uses the law prohibits. Neither does the lawyer.
For two years, the company’s software has been labeling employees’ moods. The person responsible for the AI inventory learns about it from a call center supervisor, in a meeting about call center performance.
In March 2018, the WIRED website reported that MetLife was using an artificial intelligence computer programme, Cogito, to analyse the voices of call centre agents. If an agent sounded tired during a call, a coffee cup might appear on the screen. One supervisor described it as “a cute little coffee cup” and said that the team found it useful.

In May 2026, The Atlantic reported that MetLife was using a programme to analyse its agents’ pitch and tone. Burger King was also testing a headset chatbot named “Patty,” which evaluated interactions for friendliness.
The above reports relate to software which analyses voices and supports coaching. They do not document the unannounced rollout or the management failures within Carla’s imagined firm. This article does not claim that the firms mentioned above have acted unlawfully.
A label such as “sad or distressed” is the machine’s interpretation, not proof of how the person feels.
There are already restrictions in the European Union on certain uses of AI to infer emotions at work. Since February 2, 2025, under the EU AI Act, AI-based emotion inference at work from biometric data, such as facial images, has been prohibited, with exceptions for medical or safety purposes. Commission guidance gives the example of a call centre using webcams and voice recognition to track employees’ emotions as a prohibited practice.
California’s law is phrased differently and does not have the same scope. Governor Newsom signed AB 1883 on September 30, 2026, and his signing statement asked the legislature to clarify its scope.
First, the problem for the employer: the law prohibits certain uses of software, but a list of product names does not tell the employer whether those uses are occurring.
If the company is already using some software and a feature is added to it, an inventory built only from purchase orders may miss it. There may be no separate purchase order for the feature, although it might appear in the release notes or on a settings page.
Even finding the feature is only the beginning.
Genesys, a software provider for call centres, mentions a tool in its documentation named “agent empathy analysis.” This analyses the language agents use with customers and returns a score from minus 100 to plus 100.
Genesys calls this a measure of empathy conveyed in the conversation, which is different from the agent’s actual emotional state. The language an agent uses towards a customer needs to be distinguished from how the agent feels.
Each specific use needs to be examined under California law to determine whether it is covered. That means examining how a particular feature functions and what the employer uses it for. But someone must identify the feature and establish what it does for that assessment to be useful.
The day after the signing, Fisher Phillips issued guidance to employers to review call centre, collaboration, video meeting, HR and wellbeing platforms for AI features tagged “sentiment,” “mood,” “stress” or “burnout.”
The firm also recommended writing to vendors with questions and considering contract terms requiring notice before a new feature of this kind is turned on.
One more question: how would the company learn that the vendor had added the capability, even if it had not yet been activated?
Raj’s company did not purchase an emotion-analysis solution. Instead, it purchased a phone system, and the decision to enable the feature was made years afterwards, on a settings page.
Fisher Phillips also comments that the law targets use and recommends recording that these features are turned off.
This resolves part of Raj’s issue. It does not answer who could turn the feature back on, whether anyone would notice, or whether a similar feature is active in another product.
Just because the feature was off on January 1 does not mean that it remained off on March 1.
If a company is to consider the issue closed, it needs to be able to answer the following three questions.
Which active features infer a worker’s emotional state, and what evidence shows how they work?
Who enabled each feature, and where was that decision recorded?
Who is responsible for the setting currently, and who can change it? How would the company know if it changed?

You too can ask a form of the first question. Ask your manager or HR whether workplace software uses your voice, appearance or writing to infer how you feel. Even outside California, it is a worthwhile question to ask.
Carla found out from an amber bar while her mother was in hospital. Employees should not have to wait for one to find out.
Carla and Raj’s case is entirely imaginary. It illustrates how an employee-monitoring feature could escape an AI inventory. This article is not legal advice.
Sources: AB 1883, California Legislature. Fisher Phillips, “California Bans AI Tools That Read Employees’ Emotions or Nervous System Data: FAQs for Employers,” October 1, 2026. Morgan Lewis, “California Enacts New Laws Governing Employer Use of AI,” October 9, 2026. Bloomberg Government, reporting on Newsom’s signing statement. Freshfields, analysis of the EU prohibition’s biometric scope. Wilson Sonsini, “EU Commission Issues Guidelines on Prohibited AI Practices Under EU AI Act,” February 11, 2025. Genesys Cloud, “Understand agent empathy analysis.” Tom Simonite, “This Call May Be Monitored for Tone and Emotion,” WIRED, March 19, 2018. Ellen Cushing, “The Rise of Emotional Surveillance,” The Atlantic, May 3, 2026.