What happened

On 30 September 2026 Governor Gavin Newsom signed thirteen AI bills in a single package. Four of them deal with the workplace, and together they set unusually detailed rules on how employers may use AI on their own staff:

  • SB 947 bars employers from relying solely on an automated decision system to discipline or dismiss a worker.
  • SB 951 requires layoff notices to say when AI is the reason jobs are going.
  • AB 1331 bans workplace surveillance in bathrooms.
  • AB 1883 bans AI tools that read employees' emotions or collect their neural data.

All four were approved and chaptered the same day: SB 947 as Chapter 859, SB 951 as Chapter 860, AB 1331 as Chapter 851 and AB 1883 as Chapter 853 of the Statutes of 2026. None was vetoed. The Governor's release calls them "first-in-the-nation" worker protections, summarises SB 947 as ensuring that "real people review automated employment actions", and quotes the Governor: "AI should expand opportunity – not come at the expense of workers and families."

Below is what the chaptered texts actually say, which in places differs from early summaries, followed by a comparison with EU, Turkish and Korean law.

SB 947: no dismissal by algorithm alone

SB 947 adds a new Part 5.5.5, "Automated Decision Systems in the Workplace", to the California Labor Code (sections 1520 to 1526.7). It becomes operative on 1 July 2027.

Scope. An "automated decision system" (ADS) 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". It must be used to assist or replace human discretionary decision-making and must materially affect natural persons. Spam filters, firewalls, antivirus software, calculators, databases and similar tools are excluded. "Employer" covers anyone who directly or indirectly, or through an agent, employs or controls wages, hours or working conditions. That includes labour contractors and public bodies. There is no minimum headcount.

Three outright prohibitions. Under section 1522(a), an employer may not use an ADS to:

  • prevent compliance with, or violate, labour, safety, employment or civil-rights law;
  • infer an employee's protected status under California's anti-discrimination statute; or
  • predict that a worker will exercise legal rights and take adverse action against them for it.

The core rule, in two tiers. Section 1522(b)(1) says an employer "shall not rely solely on an ADS when making a disciplinary or termination decision". That is the "no AI-only firing" headline. The second tier matters more:

  • Where an employer primarily relies on ADS output, a human must corroborate the decision. The reviewer must use the data behind the output or other relevant information. The statute gives examples: supervisory evaluations, personnel files, work product, peer reviews and witness interviews, which "may include relevant online customer reviews".
  • If the output cannot be corroborated, or the reviewer finds it inaccurate, incomplete or misleading, the employer may not use it for the decision (section 1522(c)).

Transparency after the decision. Where the employer primarily relied on an ADS, the employee may request "a meaningful, objective description" of their own data that the system used. Other people's personal information in that description must be anonymised. The employer must also give the employee a written post-use notice when it tells them of the decision. The notice must be in plain language, as a separate communication, in the language the workplace normally uses. It must state:

  • that an ADS was primarily relied on;
  • that a human reviewed and corroborated the output;
  • whom the employee can contact; and
  • that retaliation is prohibited.

Enforcement. The Labor Commissioner may investigate, order temporary relief and issue citations. Public prosecutors may sue. The civil penalty is $500 per violation. Retaliation against employees who exercise their rights under the Part is prohibited (section 1526). Section 1526.1(c) shifts the burden of proof: once it is shown that an ADS was used, the employer must prove either that it did not primarily rely on the system or that it complied with the corroboration and notice rules.

Carve-outs.

  • Collective agreements. An agreement displaces the statute only if it waives it "in clear and unambiguous terms", expressly provides for wages, working conditions and other terms of work, and itself "provides protection from algorithmic management".
  • Aerospace and defence. Uses required by, or reasonably necessary to comply with, a federal statute, regulation or binding federal contract on aircraft development or national-security, military, space or defence work are exempt.
  • Local ordinances. Ordinances that give workers more protection are not pre-empted.
  • Notice safe harbour. Complying with SB 947's notice rules satisfies substantially similar ADS notice duties under other state laws. The exception is businesses subject to the California Consumer Privacy Act, which remain bound by the automated decision-making regulations of the California Privacy Protection Agency.

SB 951: when AI is the reason for the layoff

SB 951 works through an existing statute, the California Worker Adjustment and Retraining Notification Act (Cal/WARN). Cal/WARN already requires 60 days' written notice before a mass layoff, relocation or termination at a "covered establishment", meaning a facility with 75 or more employees in the past 12 months. The notice goes to affected employees, the Employment Development Department and local officials.

The bill amends Labor Code section 1401. Where the event is "caused in whole or in substantial part by an AI system or other automated technology replacing or automating employment positions", the notice must also state:

  • the number, classification or occupation, and work location of the affected jobs;
  • the job functions that will be automated;
  • "the specific category or type" of AI system or automating technology involved; and
  • at the top, the sentence "This notice is for a technology displacement".

The Department must publish summaries of these notices, including a quarterly statewide summary. By 1 January 2028 it must also report to the Legislature on AI's effects on hiring, including "technological cessation in hiring" (the permanent end of hiring for a role because of AI). The report may include recommendations on a public notice duty for such cessation. The reporting provision is repealed on 1 January 2029.

SB 951 prohibits nothing: an employer may still replace workers with AI, as long as it says so. It also applies only where Cal/WARN's thresholds are met. Defective notices fall under Cal/WARN's existing remedies, which include a civil penalty of up to $500 for each day of violation. The bill has no urgency clause and no delayed operative date, so it takes effect on 1 January 2027 under California's general rule.

AB 1331 and AB 1883: limits on surveillance

AB 1331 defines a "workplace surveillance tool" broadly. It covers any system or device that collects employee data, communications, biometrics or behaviour other than by direct human observation, including video, audio, continuous time-tracking and geolocation. The bill prohibits using such a tool to monitor employees in a workplace bathroom unless a court orders it.

  • Employees may leave surveillance devices behind when they enter a bathroom, unless law requires them to stay available during breaks.
  • Employers may still log a single entry and exit.
  • Badges and personal alarms that a policy requires for identification, safety or access to a locked bathroom are permitted, provided they record no audio or video and have no embedded or attached AI.

AB 1883 is narrower and more novel. An employer "shall not use a workplace surveillance tool that uses artificial intelligence on employees" for either of two purposes:

  • to recognise, or make inferences or predictions about, an individual's emotional state; or
  • to collect neural data, meaning information generated by measuring the activity of the nervous system.

A clause preserving use "to ensure safety" leaves open whether a safety purpose can justify emotion inference.

Both bills are enforced by the Labor Commissioner and public prosecutors, with penalties of up to $500 per violation. Under AB 1883, an employee can also recover the penalty. Both take effect on 1 January 2027. The Governor signed them with a joint signing message asking the Legislature to fix them. He wrote that AB 1331 "may interfere with systems used by employees to access locked facilities", and that AB 1883 "lacks key definitions, which may create confusion about which tools are covered".

A note on AB 2713. Early summaries included AB 2713 (Wicks, Chapter 856) as a ban on removing digital watermarks. It is not a workplace measure. The chaptered text amends the California AI Transparency Act provision on large online platforms. Those platforms must not, "to the extent technically feasible, knowingly strip" standards-compliant provenance data or digital signatures from content. The rule applies from 1 January 2027 and binds platforms only. It does not create a general offence of removing watermarks.

Reading the package

SB 947 is a redrafted version of a bill the Governor vetoed. In October 2025 he returned SB 7, the "No Robo Bosses Act", also by Senator McNerney, without signature. His veto message made three objections:

  • it imposed "unfocused notification requirements on any business using even the most innocuous tools";
  • it restricted the use of customer ratings; and
  • the Privacy Protection Agency's forthcoming regulations should be assessed first.

SB 947 answers each one: no pre-use notice, customer reviews accepted as corroborating evidence, and the Agency's regulations preserved.

The key word is "primarily", not "solely". Few employers dismiss anyone by fully automated means, so the sole-reliance ban changes little on its own. The duty that will affect daily practice is corroboration where reliance is primary. A manager signing off on a recommendation is not enough: the reviewer must test the output against evidence and be free to reject it, and under the burden-shifting rule the employer must prove that this happened.

The drafting has rough edges.

  • The burden-shifting clause refers to a "disciplinary or deactivation decision", where the rest of the Part says "termination". "Deactivation" is the vocabulary of platform work and of SB 7.
  • The remedies clause allows injunctive relief "including punitive damages", an unusual combination.
  • No express private right of action is created.

The regimes overlap. From 2027 a California employer using AI in HR may face SB 947, the Privacy Protection Agency's automated decision-making regulations (in effect since 1 January 2026, according to the Agency, with their automated decision-making requirements to be met from 1 January 2027), anti-discrimination law and local ordinances. The safe harbour removes duplicate notices. It does not merge these regimes.

Reach. The statutes protect employees in California. Because "employer" includes anyone controlling working conditions "directly or indirectly, or through an agent", a parent company running a global HR platform from abroad is not obviously outside them. Cal/WARN expressly treats a parent as the employer at its subsidiary's establishments.

How the rules compare: EU, Türkiye, Korea

European Union. The EU regulates the same ground through product-style rules rather than labour law:

  • Annex III, point 4(b) of the AI Act classifies as high-risk any AI system intended "to make decisions affecting terms of work-related relationships, the promotion or termination of work-related contractual relationships", to allocate tasks, or to monitor and evaluate workers' performance and behaviour.
  • Deployers must assign human oversight. Under Article 26(7), employers must inform workers' representatives and affected workers before using such a system at the workplace.
  • Following the Digital Omnibus on AI, Regulation (EU) 2026/1744 (published in the Official Journal on 24 July 2026), the Annex III obligations, including these deployer duties, apply from 2 December 2027.
  • On emotion recognition the EU goes further than AB 1883. Article 5(1)(f) has prohibited AI that infers emotions in the workplace since 2 February 2025, except for medical or safety reasons.
  • For platform workers, Article 10(5) of the Platform Work Directive requires decisions to restrict, suspend or terminate an account or contract to be "taken by a human being". It must be transposed by 2 December 2026.

The EU acts before deployment, through notice, conformity duties and oversight design. California acts at the moment of decision, through corroboration, post-use notice and a reversed burden of proof. A multinational needs both.

Türkiye. Türkiye has no AI-specific employment statute, but existing law reaches much of the same conduct:

  • Data protection. Article 11(1)(g) of the Personal Data Protection Law No. 6698 gives every data subject the right to object to an outcome against them produced by analysing their data "exclusively through automated systems".
  • Valid reason. Under Article 18 of the Labour Law No. 4857, employees with job security (30 or more workers at the workplace and six months' service) may be dismissed only for a valid reason.
  • Notice and defence. Article 19 requires written notice stating the reason "clearly and precisely". It also requires the employee's defence to be heard before a dismissal for conduct or performance.
  • Burden of proof. Article 20 places the burden of proving the valid reason on the employer.
  • Collective dismissals. Article 29 requires 30 days' notice to union representatives, the regional labour directorate and the Turkish Employment Agency, stating the reasons. The Article lists "technological" reasons among the grounds for a collective dismissal, but unlike SB 951 it does not require the employer to say that AI is the cause.

Together, these rules make a dismissal resting on an unexplained algorithmic score hard to defend before a Turkish labour court.

Korea. Korean law leads to a similar place:

  • Automated decisions. Article 37-2 of the Personal Information Protection Act, in force since 15 March 2024, lets data subjects refuse a decision made by a fully automated system, AI included, that significantly affects their rights or obligations. They may also ask for an explanation.
  • The contract exception. The refusal right does not apply where the decision rests on consent, a legal basis or the performance of a contract (Article 15(1) items 1, 2 and 4). That covers much of employment. The explanation right survives, and the criteria and procedures of automated decisions must be published.
  • Dismissal rules. The Labor Standards Act requires just cause (Article 23) and written notice of the reason and date (Article 27). Managerial dismissals require 50 days' notice to, and good-faith consultation with, the union or worker representatives (Article 24).
  • Framework Act on AI. In force since 22 January 2026, it lists judgments or evaluations "such as hiring" that significantly affect individual rights among the areas where AI may be "high-impact".

A just-cause or valid-reason standard combined with a right to object or refuse already requires a human who can explain the decision. California adds the procedure for doing and recording that review.

What multinational employers should do

Turkish and Korean groups with staff in California should work to two dates: 1 January 2027 and 1 July 2027.

  • Inventory the tools that touch discipline and dismissal. Include performance scoring, productivity monitoring, attendance flags and "risk" dashboards, not only tools marketed as AI. Global HR systems run from headquarters count if they are used on California staff.
  • Classify each decision path as sole, primary or supporting reliance. Remove sole-reliance paths. For primary-reliance paths, write a corroboration procedure that names the reviewer, lists the evidence to be checked and records the outcome, including any rejection of the output.
  • Prepare the post-use notice and the data-description process in advance, in the languages the California workforce actually uses, with anonymisation of third-party data built in.
  • Add a technology-displacement check to the Cal/WARN template. Board papers and investor messaging that credit AI with headcount savings will be read alongside the notice.
  • Audit wearables, cameras and monitoring software for bathroom coverage, emotion inference and neural-data collection. That includes fatigue headsets, call sentiment analysis and "engagement" scoring.
  • Map SB 947 onto the Privacy Protection Agency's rules rather than running two programmes.
  • Use one global standard where possible. A group already complying with KVKK Article 11, PIPA Article 37-2 and the AI Act's worker-information duty can usually extend a single human-review and explanation procedure to California.

What to watch next

  • Clean-up bills. The Governor has asked for amendments to AB 1331 and AB 1883.
  • Guidance. Watch for any Labor Commissioner guidance or forms for SB 947 notices before 1 July 2027.
  • Displacement data. The Employment Development Department's displacement summaries from 2027 will be the first official data on AI-driven layoffs in any US state.
  • The January 2028 report on "technological cessation in hiring", which may lead to a second notice duty, and the EU's Annex III start date of 2 December 2027.

Frequently asked questions

Does SB 947 ban AI in HR decisions?

No. It bans relying solely on an ADS for discipline or termination, and it bans three specific uses. Primary reliance remains lawful if a human corroborates the output and the employee receives a post-use notice and, on request, a description of their data.

We are a Turkish company with five employees in California. Are we covered?

SB 947, AB 1331 and AB 1883 set no minimum employer size. SB 951 applies only where Cal/WARN does: establishments of 75 or more and layoffs above its thresholds.

Can a collective agreement opt out of SB 947?

Only an agreement that meets three conditions: it waives the Part in clear and unambiguous terms, expressly provides for wages, working conditions and other terms of work, and itself provides protection from algorithmic management.

When does each law apply?

AB 1331, AB 1883, SB 951 and AB 2713 apply from 1 January 2027. SB 947 applies from 1 July 2027.

Burhan Doğuş Ayparlar's View

This section sets out my personal assessment as the founder of this site and an attorney at law (Türkiye).

The headline, "no AI-only firing", undersells the statute and slightly misleads. Almost no one is fired purely by machine. What happens in practice is that a dashboard flags someone, a manager without the time or data to question it adds a signature, and the decision is recorded as human. SB 947's real achievement is to treat that pattern as primary reliance and attach consequences to it. A human must actually corroborate the output, the employee must be told, and in a dispute the employer must prove the review happened. That is the right target, and I expect it to matter far more than the sole-reliance ban.

The legislative history is just as instructive. The broader version was vetoed for burdening "even the most innocuous tools"; the version that passed tied its duties to the decision rather than to the technology. Turkish and Korean legislators working on AI bills could take that lesson.

From the Turkish side, I do not think California is asking for anything Turkish law does not already require in substance. A dismissal must rest on a valid reason stated clearly and precisely, the employer bears the burden of proving it, and KVKK gives the employee a right to object to a solely automated outcome. An employer that cannot explain why its system flagged someone will lose in Ankara as surely as in Sacramento. What California adds is procedure: written corroboration, a stand-alone notice and a data description. Turkish employers would do well to adopt that procedure across the group, not only for California payroll.

SB 951 deserves more attention than it has received. It prohibits nothing, but it makes companies state in a public filing that AI took the jobs. Companies that tell investors AI is driving efficiency and tell staff that layoffs reflect "restructuring" will now have to reconcile the two in California. I expect the resulting data to shape the debate in Türkiye and Korea too, where AI's effect on employment is still argued with very little evidence.

Finally, banning emotion inference and neural-data collection at work is right; the EU got there first. But when the Governor signs two bills while saying that one lacks key definitions, the sensible course is to comply with the clear core now and watch for the amendments.

This article is for information only and does not constitute legal advice. It is based on the chaptered texts of SB 947 (Ch. 859), SB 951 (Ch. 860), AB 1331 (Ch. 851), AB 1883 (Ch. 853) and AB 2713 (Ch. 856), Statutes of 2026, as published by the California Legislature; the Governor's release and joint signing message of 30 September 2026; and the October 2025 veto message on SB 7. The 2 December 2027 date for EU Annex III obligations is taken from Regulation (EU) 2026/1744 (Digital Omnibus on AI) as published in the Official Journal on 24 July 2026. For bills without an express operative date, the effective date stated applies California's general 1 January rule. Statements about Turkish and Korean law are general; specific cases require individual assessment. The analysis and assessments are the author's own.