California has written four workplace AI laws that all land on the same desk. Whoever in an employer signs off on a firing, files a layoff notice or buys a monitoring tool now has statutory duties that did not exist last week, and HR is the function most likely to carry them.
Governor Gavin Newsom signed 13 AI bills, according to his office’s September 30 release. Four of them reach employment: SB 947, SB 951, AB 1331 and AB 1883. They cover how employers discipline and fire people with software, how they disclose layoffs caused by automation, and where and how they may watch employees. This piece reads the bill text on the Legislature’s site rather than summaries of it, because the details that matter to an HR team sit in definitions, exceptions and enforcement clauses.
The Governor’s framing, and the labor movement’s
The release groups the four bills as worker protections. Newsom said “AI should expand opportunity”, and that it should not come at the expense of workers and families. The release also carries a statement from a labor leader. “By signing these four first-in-the-nation bills, Governor Newsom has shown the country how we can begin to put guardrails on AI at work,” said Lorena Gonzalez, President of the California Federation of Labor Unions, AFL-CIO.
Whether any of this is first in the nation is the sponsors’ claim, and this article does not test it. What the statutes do can be read directly.
SB 947: a human has to check the machine before someone is disciplined or fired
SB 947 adds a new part to the Labor Code on automated decision systems in the workplace. Its definition is broad. An automated decision system, or ADS, is any computational process derived from machine learning, statistical modeling, data analytics or artificial intelligence that issues a score, classification or recommendation used to assist or replace human decisions that materially affect people. The text carves out spam filters, firewalls, antivirus software, identity and access management tools, calculators, databases and datasets. A performance-scoring module inside an HCM suite, a productivity ranking in a workforce management tool or a flight-risk model does not appear on that exclusion list.
The operative rules sit in two layers.
Layer one: what an ADS may never be used to do
An employer may not use an ADS to prevent compliance with, or violate, any federal, state or local labor, safety, employment or civil rights law. It may not use one to infer an employee’s protected status. And it may not use one to predict a worker’s exercise of legal rights and then take adverse action against that worker. The third item matters for anyone running attrition or models that predict who will complain, because the bill lists predicting rights-exercise and acting against the worker as one prohibited use.
Layer two: no sole reliance on an ADS for discipline or termination
An employer may not rely solely on an ADS when making a disciplinary or termination decision. Where the employer primarily relies on an ADS output, it must direct a human to corroborate the decision. The statute lists the kinds of material that count as corroboration: supervisory or managerial evaluations, personnel files, employees’ work product, peer reviews and witness interviews, which may include relevant online customer reviews. If the employer cannot corroborate the output, or the reviewer concludes it is inaccurate, incomplete or misleading, the employer may not use it for the decision.
That second sentence is the one that changes workflows. A reviewer who clicks “approve” on a system recommendation is not corroborating anything. The record has to show that a person looked at independent evidence and reached a view.
The notice and the data request
When an employer primarily used an ADS for a disciplinary or termination decision, it must give the affected employee a written post-use notice at the time it communicates the decision. The notice has to be a separate, stand-alone communication in plain language, in the language the employer normally uses with its workforce, delivered by a simple method such as an email or hyperlink. It must say four things: that the employer primarily relied on an ADS, that a human reviewed the decision and corroborated the output, who the employee can contact about it, and that retaliation for exercising these rights is prohibited. The employee can also request a meaningful, objective description of their own data that the ADS used. The employer must anonymize other people’s personal information in that description.
Enforcement and the burden of proof
The Labor Commissioner can investigate, order temporary relief and issue citations, and public prosecutors can enforce the part. The civil penalty is $500 per violation. One clause deserves a second read. In a proceeding over sole reliance or failed corroboration, once it is shown that an ADS was used for the decision, the employer must demonstrate that it did not primarily rely on the ADS or that it complied with the corroboration and notice requirements. The employer, in other words, carries the burden of showing its process. An HR team that cannot produce the reviewer’s notes, the evidence consulted and the notice sent is in a weak position the moment a complaint arrives.
The bill also reaches beyond direct employers. Its definition of employer includes labor contractors and farm and foreign labor contractors, and it covers public employers including cities, counties, school districts and state universities.
SB 951: layoffs caused by automation get their own label
SB 951 amends California’s state mass-layoff notice law. The base rules in the text are familiar: covered establishments employ 75 or more people, a mass layoff is 50 or more employees in a 30-day period, and the employer must give written notice 60 days before the order takes effect to affected employees, the Employment Development Department, the local workforce development board and local elected officials.
The new part applies when a mass layoff, relocation or termination is caused in whole or in substantial part by an AI system or other automated technology replacing or automating positions. The notice must then also state the number, classification or occupation and work location of the layoffs substantially due to the automation, the job functions that will be automated, and the specific category or type of AI system or automating technology involved. The statute prescribes a header. The notice must carry the statement “This notice is for a technology displacement” at the top.
The Employment Development Department must publish a summary of these notices as part of its regular reporting and post a quarterly statewide summary of technology displacements. That makes the HR team’s description of the cause a public record. The words chosen in a layoff notice, and the decision about whether automation was a substantial cause, will be public.
A separate section addresses a harder problem. The bill defines a “technological cessation in hiring” as the permanent ending of hiring or contracting for a particular occupation caused in whole or in part by AI, whether or not current workers in that role remain. It does not impose a notice duty for that situation. It instead directs the Employment Development Department to report to the Legislature by January 1, 2028 on AI’s effects on hiring, including recommendations on whether employers should have to give public notice of such cessations. Talent acquisition leaders who have frozen requisitions for roles they expect software to absorb should read that section as a signal about where the disclosure debate is heading.
AB 1883 and AB 1331: what employers may not monitor
The two surveillance bills share a definition of a workplace surveillance tool: any system, application, instrument or device that collects or facilitates collection of employee data, activities, communications, actions, biometrics or behaviors by means other than direct observation by a person. The list of examples runs from video and audio surveillance to continuous time-tracking tools, geolocation and electromagnetic tracking.
AB 1883: no emotion recognition, no neural data
AB 1883 bars employers from using an AI-enabled workplace surveillance tool on employees to recognize, or make inferences or predictions about, an individual’s emotional state, or to collect neural data. Neural data is defined as information generated by measuring the activity of an employee’s central or peripheral nervous system that is not inferred from non-neural information. The bill does not stop an employer from using a surveillance tool for safety or for any purpose outside those two prohibited uses. It also contains exemptions for tools used in developing aircraft for the national airspace and in developing products or services for national security, military, space or defense purposes.
For HR technology buyers the practical task is a product review. Engagement analytics, interview-assessment software and call-center coaching tools that score sentiment or tone from voice or video would need a hard look against the emotional-state language. The question to put to each vendor is what the system infers and from which signals, because the statute turns on inference as well as direct recognition.
AB 1331: the bathroom rule
AB 1331 prohibits employers from using a workplace surveillance tool to monitor or surveil employees in a workplace bathroom, unless a court order directs otherwise. Employees have a right to leave surveillance tools on their person behind when they enter a bathroom, except where they must remain available during meal or rest periods under federal or existing state law. An employer may check tools for the one-time entry and exit. An employer does not violate the rule when an employee voluntarily brings a tool in, or brings in a badge or personal alarm because company policy requires it for identification or safety. The text also exempts certain tools that do not record audio or video and have no physically embedded AI.
Both bills use the same enforcement design as SB 947: Labor Commissioner citations and civil actions, enforcement by public prosecutors, injunctive relief and a penalty of up to $500 per violation.
What it means for the HR leader
The four laws share one structural idea: a person has to be identifiable, in the record, at the point where software touches an employee’s job or privacy. Three practical consequences follow.
Inventory first. The ADS definition is wide enough that an HR system’s scoring feature, a workforce management tool’s flagging logic and a vendor’s “risk” signal may all qualify. Our earlier column on inventorying hiring AI before the year ends makes the same case for hiring tools; the California text extends the list to discipline and termination tools.
Build the corroboration file now. Because the employer carries the burden once an ADS is shown to have been used, the reviewer’s evidence matters as much as the decision. A template that records what was consulted, who reviewed it and why the output was accepted or rejected costs little to build and is the thing a Labor Commissioner investigator will ask for.
Treat the vendor contract as a compliance document. Employers need to know whether a tool infers emotion, how a scoring model uses employee data, and whether the vendor can produce the data description an employee may request. Those answers belong in procurement, not in a post-complaint scramble.
The broader legal frame is also getting less forgiving of the idea that software changes the rules. We argued in There Is No AI Exception in Employment Law that existing anti-discrimination duties already apply to automated tools. California has now added a layer that is specific to automation, with its own notices, burdens of proof and penalties.
Where employers should be careful about the text
The sections above describe the bill text published on the Legislature’s site. Two cautions apply. First, compliance dates, regulations and agency guidance are separate from the statutory text, and an employer should confirm effective dates and any implementing rules with counsel before changing a process. Second, several terms leave judgment calls that this article does not resolve, including when an employer “primarily” relies on an ADS output, and when automation is a “substantial” cause of a layoff. Those are the questions that will be argued in citations and lawsuits, and the employers who document their reasoning early will be better placed than those who improvise later.
Source: Office of Governor Gavin Newsom. Bill text: SB 947, SB 951, AB 1883, AB 1331.
