Gov. Gavin Newsom signed California's No Robo Bosses Act on Sept. 30, the final day he had to act on the 2025-26 session's bills, and the state's description of it holds up: an employer may no longer fire or discipline a worker on an automated system's say-so. Senate Bill 947, by Sen. Jerry McNerney of Pleasanton, requires a person to corroborate the machine's output, requires a written notice telling the worker that an automated system was primarily behind the decision, and gives the state labor commissioner a $500 civil penalty to collect for each violation.
The law becomes operative on July 1, 2027. What it regulates is the form of the decision rather than the quality of the system that produced it. Nothing in the text says the corroborating human must be independent of the decision, trained in how the system works, or able to reverse it. That person is the employer's to choose, and the statute permits the employer to point them at the same data the system used.
"No worker should ever be fired or disciplined by a robo boss," McNerney said in a statement after the signing, calling the law the first of its kind in the country.
It is one of four labor bills Newsom signed that day. The others require notice when a mass layoff is caused by an AI system, bar employer surveillance tools in workplace bathrooms, and prohibit employers from predicting a worker's emotional state or collecting brain data. The package was sponsored by the California Federation of Labor Unions, AFL-CIO, one year after Newsom vetoed the predecessor of SB 947. Its president, Lorena Gonzalez, described the bills as a way to "put guardrails on AI at work to ensure human oversight over this technology."
What the law tells an employer to do
The act adds Part 5.5.5 to the Labor Code, and it starts at Section 1520 with definitions. An automated decision system is "any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that issues simplified output, including a score, classification, or recommendation, that is used to assist or replace human discretionary decisionmaking and materially impacts natural persons." The same section excludes email spam filters, firewalls, antivirus software, access management tools, a calculator and a database.
Section 1522 also bans three uses outright: an automated system may not be used in a way that prevents compliance with or violates labor, safety, employment or civil rights law; may not be used to infer an employee's protected status; and may not be used to predict and then act against a worker for exercising legal rights.
The firing provision is narrower than the bill's name suggests. An employer may not rely solely on an automated system when making a disciplinary or termination decision, and when an employer primarily relies on one of these outputs, it has to "direct a human to corroborate the decision." The corroboration may draw on "data that was collected or used to produce the ADS output or other relevant corroborating or supporting information," and the list of examples is supervisory or managerial evaluations, personnel files, work product of employees, peer reviews, and witness interviews, which the text says may include relevant online customer reviews. If the output cannot be corroborated, or the reviewer concludes it is inaccurate, incomplete or misleading, the employer cannot use it to make the decision.
The corroborating human is the employer's choice
Two terms decide when the duty applies and what satisfies it, and the part defines neither. "Primarily relies" sets the threshold: an employer that used a system as one input among many is untouched, and an employer that primarily relied on it owes corroboration, but nothing in the text says where that line falls. "Corroborate" is the duty itself, and it is undefined too, which leaves open whether a second look at the same inputs counts.
The list of acceptable corroboration is explicitly non-exhaustive, and it opens with supervisory or managerial evaluations. That is the category of evidence an employer builds by having managers write down what they think of their reports, and the category these systems displace when they rank or flag those same workers. The statute allows the output and the paperwork to agree, and it allows the manager who wanted the review to conduct it.
Section 1522(c) gives the human one direction of authority. A reviewer who concludes the output is inaccurate, incomplete or misleading stops the employer from using it. A reviewer who concludes otherwise has satisfied the statute, as long as corroboration of some kind exists. Nothing requires that the reviewer be told they may override the system, be trained on it, or that the employer record why the corroboration was enough.
Utah's prior authorization law bans nothing and reaches the same question from the other side, putting a name on the reviewer who signs a denial and requiring insurers to disclose their use of AI, an arrangement we described when it passed. What California adds is the duty to produce a second look before the decision takes effect, and the discretion to decide who supplies it.
The notice says a human agreed, not what the machine concluded
Section 1524 requires a written postuse notice, delivered at the time the employer tells the employee about the decision. It has to be plain language in a stand-alone communication, in the language the employer uses for routine communications, delivered by a simple method such as email or a hyperlink. It has to say that the employer primarily relied on an automated decision system, that a human reviewed the decision and corroborated the output, whom the employee can contact with questions about the decision and the right to see a description of their own data, and that retaliation for exercising rights under the part is prohibited.
What the notice does not have to contain is the output. An employee learns that a system was primarily behind the decision but not the score, the ranking, the threshold it crossed, or the company that supplied it. The separate data right in Section 1522(d) covers "a meaningful, objective description of the employee's own data" that the system used, and Section 1522(e) requires the employer to anonymize the personal information of customers, other employees and other individuals when it supplies that description. Because the notice is due at the same moment as the decision, it works as a record for what follows, a complaint to the labor commissioner or an unemployment appeal, not as a warning the worker can act on.
Enforcement is public, and the price is five hundred dollars
The part hands enforcement to the state. The labor commissioner can investigate, order temporary relief, issue a citation against an employer, and file a civil action under the procedures that already govern wage citations, and a public prosecutor can sue as well. In either forum, Section 1526.1(c) shifts the burden once it has been demonstrated that an automated system was used to make the decision: the employer then has to show that it did not primarily rely on the system or that it complied with the corroboration and notice sections. The part also bars retaliation against an employee who files a complaint or cooperates in an investigation, and it does not preempt city or county ordinances that give employees equal or greater protection.
An employer that violates the part is subject to a civil penalty of $500 per violation, and a court action can also seek injunctive relief, punitive damages, and attorney's fees and costs. Workers cannot enforce it themselves. CalMatters reported that the bill lost a private right to compel an employer to follow the law, an appeals process for workers, and coverage of contractors as it moved through the Legislature.
One clause reaches further than the rest of the part. The burden shift in Section 1526.1(c) applies "once it has been demonstrated that an ADS was used to make a disciplinary or deactivation decision." Deactivation appears in no other section: not the prohibitions, not the notice, and not the definitions, which cover scores, classifications and recommendations but never say what it means to deactivate a worker. Read literally, the clause contemplates a decision that switches a person off, the vocabulary of platform work rather than employment, and it leaves the protections built around discipline and termination unclear where the rest of the part may not reach. Whether the word is a drafting error or a deliberate hook will be for the labor commissioner or a court.
Contractors, union waivers, and who is covered
Coverage depends on being an employee, and the definition is broad in some directions. An employer includes state and local government, and it includes labor contractors, farm labor contractors and foreign labor contractors, which means a worker supplied by a staffing agency is covered even though the agency rather than the worksite runs the payroll. Independent contractors are not covered. That leaves out a growing share of the workforce scored and dispatched by software, and it leaves the corroboration duty resting on an employment relationship that platform business models are built to avoid.
Section 1526.5 lets a union contract waive the part entirely, provided the agreement does so "in clear and unambiguous terms," sets wages and working conditions, and provides "protection from algorithmic management." Gonzalez's argument for the package was that most workers are not in a union and so have no seat at the table when their employer decides how to use the technology. The statute she sponsored then allows the opposite arrangement for the workers who do: a bargaining unit can trade the default rule for terms its members negotiated, while workers without a contract keep the statutory version, enforced only by a state agency with a $500 citation. The part steps aside where a federal law or binding federal contract requires an automated system for aircraft development or for national security, military, space or defense work, and it relieves an employer that complies with the notice section of substantially similar notice duties under other state laws, except for the California Consumer Privacy Act's automated decisionmaking rules.
A judgment for clinicians and a veto for the protection behind it
The same day showed where the governor drew his line, and it was in health care rather than the workplace. Newsom signed AB 1979, which says health care professionals rather than AI have the final say in decisions about patient care, and he vetoed AB 2575, which would have protected health care workers from retaliation if they refused to follow an AI-generated recommendation.
The nurses' union put the contrast plainly. "That makes no sense to any nurse at the bedside," said Sandy Reding, president of the California Nurses Association, adding that a nurse who sees the algorithm is wrong and acts to save a patient can still be disciplined or fired by her employer. The governor's office says the clinicians' bill ensures that doctors and licensed providers can use their own professional judgment when these tools are part of patient care, and requires the developers of clinical decision tools to take reasonable steps to reduce known or predictable bias. A measure to prohibit replacing mental health workers with AI was not among the bills he announced signing.
The corroboration duty starts on July 1, 2027. Between now and then, the questions that will decide how much it changes are who the labor commissioner says a corroborating human has to be, how "primarily relies" gets read the first time an employer argues the decision was its own, and whether the notice shapes a decision or describes it afterward. The statute has already answered the first. The choice of human belongs to the employer, and the law asks only that one existed.
Primary sources
- California Legislature, SB 947, Employment: automated decision systems, enrolled text, Sept. 4, 2026, for the definitions in Section 1520, the employer duties and prohibitions in Section 1522, the postuse notice in Section 1524, the enforcement, burden-shift and penalty provisions in Sections 1526 and 1526.1, the collective bargaining waiver in Section 1526.5, the federal contract exception in Section 1526.6, and the July 1, 2027 operative date in Section 1526.7.
- Office of Gov. Gavin Newsom, California's nation-leading AI framework just got stronger, Sept. 30, 2026, for the list of signed bills, the description of AB 1979 and the health care provisions, and the statement from California Federation of Labor Unions president Lorena Gonzalez.
- Office of Sen. Jerry McNerney, Newsom Signs McNerney's No Robo Bosses Act of 2026, Sept. 30, 2026, for the senator's statement, the sponsorship by the California Federation of Labor Unions and the description of the law as the first of its kind.
- CalMatters, On AI, Newsom gives labor only some of what it demanded, Sept. 30, 2026, for the provisions the bill lost during the session, the AB 2575 veto, the statement from California Nurses Association president Sandy Reding, the comparison to the 2025 veto, and the polling on AI and job loss.
- California Legislature, AB 2575 bill status, for the Sept. 30, 2026 veto.