California Assembly Bill 2027, the Robot Replacement Training Prevention Act, would prohibit employers from using a worker’s personal information to train an AI system designed to replicate or replace that worker’s job. It would also bar employers from selling or transferring worker data to third parties for the same purpose. The bill, introduced by Assemblymember Coby Ward and co-sponsored by the Teamsters, cleared the Assembly Privacy and Consumer Protection Committee on 22 April and is now before the Assembly Appropriations Committee, with a suspense hearing scheduled for 14 May.

The bill creates a new regulatory category: worker data sovereignty over AI training. Existing privacy laws give workers some control over how their personal data is collected and processed. Existing employment laws give workers some protection against discrimination in automated decisions. AB 2027 sits between these: it targets the specific use of worker data to build the systems that could make workers redundant. If enacted, it would be the first US law to draw that line.

What the bill does

AB 2027’s core prohibition is narrow and specific. It would bar employers from using a worker’s personal information to train an AI system “to replicate, automate, or replace” that worker’s job. It would also prohibit selling or transferring that data to a third party for the same purpose. The bill does not ban AI in the workplace. It does not prohibit employers from using AI to augment work, improve productivity, or automate tasks. It prohibits using the worker’s own data to build the system that replaces them.

The distinction matters for governance teams. Performance data, productivity telemetry, workflow recordings, communication patterns, and task completion metrics are all routinely collected by employers. Some of that data flows into AI training datasets, often through vendor platforms that process employer data under service agreements that may or may not restrict training use. AB 2027 would require employers to know where that data goes and ensure it is not being used to train replacement systems.

The California AI cluster

AB 2027 is one of several workplace AI bills advancing through the California legislature in 2026. The Transparency Coalition’s 8 May update tracks the full cluster:

AB 1898 would require employers to provide written notice to workers before deploying workplace AI tools that assist in decisions or conduct surveillance. Employers would need to maintain an updated list of all AI tools in use and describe whether any jobs or tasks will be replaced or automated, including expected timelines. The bill includes penalties up to USD 500 per employee per violation, enforced by the Labour Commissioner. It cleared committee and was sent to Appropriations on 22 April.

SB 947, the No Robo Bosses Act of 2026, would ban employers from relying solely on automated decision-making systems to fire or discipline workers, requiring human oversight and independent verification. It would also prohibit predictive behaviour analysis systems that profile employees and take adverse action based on AI predictions. SB 947 is a reintroduction of SB 7, which passed both chambers in 2025 before Governor Newsom vetoed it.

SB 928 would specify that California State University instructors must be human. It passed the state Senate 37 to 0 on 23 April, a unanimity that signals strong legislative appetite for worker protections against AI substitution.

Lorena Gonzalez, president of the California Federation of Labor Unions, framed the legislative push directly: “Right now, there are absolutely no restrictions on how employers can use artificial intelligence to arbitrarily discipline and fire their workers. Employers are devastating workers’ livelihoods and taking no responsibility for the callous decisions of this unchecked technology.”

Why this matters beyond California

California has a track record of setting regulatory patterns that other states and countries follow. The California Consumer Privacy Act influenced GDPR-adjacent legislation worldwide. California’s pay transparency laws triggered a cascade of similar requirements across US states. If AB 2027 passes, other states with active AI legislative programmes are likely to consider similar worker-data provisions.

The Littler survey SAW covered on 15 May found 84% of US employers expect AI-related regulatory impact in the next 12 months. AB 2027 is one of the bills they should be watching. The same survey found only 54% of employers restrict the information employees can enter into AI systems. AB 2027 would impose restrictions from the other direction: not just what workers put into AI tools, but what employers take from workers to feed into AI systems.

The bill also intersects with the DOJ’s intervention in xAI’s challenge to Colorado’s SB24-205. Colorado’s law requires reasonable care to protect consumers from algorithmic discrimination. AB 2027 addresses a different harm: the use of worker data to train systems that eliminate workers. Both laws face potential federal preemption challenges under the Trump administration’s National AI Framework, which pushes for a single federal standard. Employers operating across multiple states cannot afford to wait for preemption to resolve. The safer path is to build governance that addresses both worker-data and algorithmic-discrimination requirements simultaneously.

What HR and data governance teams should do

Audit where worker data flows into AI training. Most organisations do not have a complete picture of which vendor platforms use employer-provided data for model training. HR systems, productivity platforms, communication tools, and project management software all generate worker data. If any of those vendors’ terms of service permit training use, the organisation needs to know before AB 2027 or similar legislation makes it their compliance problem.

Separate worker datasets from general training corpora. If the organisation builds or fine-tunes AI models internally, worker performance data, workflow recordings, and task metrics should be stored and governed separately from general training data. AB 2027’s prohibition is specific to data used to train replacement systems, but the operational control required to demonstrate compliance is data segregation and clear documentation of training data sources.

Review vendor contracts for training-use clauses. SaaS vendors that process worker data may include broad training-use clauses in their terms of service. AB 2027 would make the employer liable for data the vendor uses for AI training purposes if that training produces systems designed to replicate or replace worker roles. Contract review should identify these clauses and negotiate explicit carve-outs for worker data.

Build worker-facing transparency mechanisms. AB 1898’s notice requirements (written disclosure of AI tools in use, descriptions of automation timelines, updated annual lists) provide a practical template for what transparency looks like. Even if AB 1898 does not pass, the pattern of worker-facing AI disclosure is emerging across multiple jurisdictions. Building the disclosure mechanism now avoids a retrofit later.

Track the Appropriations suspense hearing. The Assembly Appropriations suspense hearing is scheduled for 14 May. Bills that survive suspense move to the full Assembly floor. AB 2027’s progress through Appropriations will signal whether California is serious about worker-data AI training restrictions or whether the bill stalls at the cost-assessment stage. Either outcome informs governance planning.

The bill has not passed yet

AB 2027 is in committee. It may be amended, stalled, or vetoed. Governor Newsom vetoed SB 7 (the predecessor to the No Robo Bosses Act) in 2025, citing overlapping protections in existing California Privacy Protection Agency regulations and CRD AI regulations that took effect in October 2025. The same veto logic could apply to AB 2027 if the governor determines existing data protection frameworks already cover the conduct the bill targets.

But the governance question does not depend on any single bill passing. Worker-data AI training restrictions are on the legislative agenda in the world’s fifth-largest economy. Multinationals with California operations or California-resident employees should treat the concept as directional: if it does not arrive through AB 2027, it will arrive through a successor bill, a regulatory action, or a court ruling. Designing governance around worker-data controls now is cheaper than retrofitting after enactment.

Sources

  • California Legislature, “AB-2027 Worker data: prohibitions: artificial intelligence,” bill status page, 2025-2026 session (bill text, committee history, sponsor). leginfo.legislature.ca.gov
  • Transparency Coalition, “AI Legislative Update: May 8, 2026,” 6 May 2026 (AB 2027 Appropriations referral, AB 1898/AB 1883/SB 928 status, suspense hearing dates, national legislative tracker). transparencycoalition.ai
  • Inland Empire Labor Council AFL-CIO, “California’s 2026 Legislative Agenda for Worker Rights,” 25 March 2026 (AB 2027 as “Robot Replacement Training Prevention Act,” Teamsters co-sponsorship, SB 947/AB 2575/SB 928 cluster). ielabor.org
  • Hard Reset Media, “Your Boss’s Algorithm is Watching. California Wants to Make It Look Away,” 28 March 2026 (AB 2027 detail, SB 947 No Robo Bosses Act, SB 951 90-day AI layoff notice, Gonzalez quote). hardresetmedia.com
  • Lexology / Troutman Pepper, “Proposed State AI Law Update: April 27, 2026,” 27 April 2026 (AB 2027 committee passage confirmation, SB 947 amendment and committee passage). lexology.com
  • California Legislature, “AB-1898 Workplace artificial intelligence tools,” bill text, 2025-2026 session (written notice requirements, AI tool list, automation timeline disclosure, penalty structure). leginfo.legislature.ca.gov