California Just Banned Employers From Letting AI Fire Workers on Its Own: Your 5-Step Compliance Plan

California Just Banned Employers From Letting AI Fire Workers on Its Own: Your 5-Step Compliance Plan

California will soon be the first state in the country to require a human to sign off before an employer can fire or discipline a worker based on what an artificial intelligence tool recommends. Governor Newsom just signed a bill informally known as the “No Robo Bosses Act” into law late yesterday after vetoing a similar measure last year. SB 947 requires a person to independently confirm the call before “primarily relying” on an AI tool to take an adverse employment action against an existing employee. Employers will also have to give the affected employee a written notice explaining the situation at the moment they deliver the news. The law takes effect July 1, 2027, which gives you plenty of time to inventory your tools and prepare. Here’s what you need to know and a five-step plan to get ready.

3 Key Things the New Law Requires

SB 947 creates three new obligations for employers:

• Cannot solely rely on AI. Beginning July 1, 2027, you cannot rely solely on an automated decision system when making a disciplinary or termination decision about an existing employee. A “system” is defined broadly as any computational process derived from machine learning, statistical modeling, data analytics, or AI that produces a simplified output such as a score, classification, or recommendation used to assist or replace human discretionary decision-making. The law carves out ordinary IT infrastructure, so spam filters, firewalls, antivirus software, identity and access management tools, calculators, databases, and datasets do not count.
• Must have a human corroborate the decision. If you “primarily rely” on an automated tool’s output to discipline or terminate an employee, you must direct a human reviewer to corroborate the decision using the data that produced the output or other relevant information. The statute offers a menu of acceptable corroborating evidence, including supervisory evaluations, personnel files, the employee’s work product, peer reviews, and witness interviews. If the reviewer cannot corroborate the output, or concludes it is inaccurate, incomplete, or misleading, you cannot use it to support the decision at all.
• Need to provide a post-use notice. At the time you inform the employee of the decision, you must provide a written notice in plain language, delivered as a separate standalone communication in whatever language you routinely use with your workforce. The notice must state that you primarily relied on an automated system, that a human reviewer corroborated the output, and that retaliation is prohibited. It also must include contact information for a human who can explain the decision. Separately, employees can request a meaningful, objective description of the last 12 months of their own data that the system used, limited to one request per year.

NOTE: The new law only applies to decisions about existing employees, not applicants.

Law Scaled Back From Original Version

The final version signed by Governor Newsom is considerably narrower than both last year’s version and the original bill introduced at the start of this session.

• Coverage now is limited to “employees” rather than the broader “workers” category, and gig worker “deactivation” language was stripped out.
• The bill’s original ban on predictive behavior analysis was also removed.
• The bill no longer requires detailed disclosures that would have forced employers to name the AI vendor and product and turn over impact assessments.
• The final law doesn’t include the express employee private right to sue that appeared in earlier drafts. However, the burden-shifting provision refers broadly to “any civil action or administrative proceeding brought pursuant to this part.” Expect plaintiffs’ lawyers to focus on that language in an attempt to expand the reach of the law.

Narrow Exemptions

• Unionized workforces can opt out through a collective bargaining agreement that explicitly waives the law in clear and unambiguous terms, covers wages and working conditions, and provides protection from algorithmic management.
• There is a narrow federal carve-out for aircraft development in the national airspace and for national security, military, space, and defense work, but only for those covered operations.

Compliance Doesn’t Automatically Satisfy ADMT Privacy Requirements

Compliance with the notice requirements will satisfy substantially similar notice obligations under other California laws, but that safe harbor does not relieve businesses covered by the CCPA from the California Privacy Protection Agency’s automated decision-making technology regulations (read more here).

Enforcement and Remedies

The Labor Commissioner may enforce the law through citations and civil actions, and public prosecutors have independent authority to bring enforcement actions. The law carries a $500 civil penalty per violation, with injunctive relief, punitive damages, and attorneys’ fees available.

Your 5-Step Compliance Plan

1. Inventory every AI tool that touches discipline. You should look into any data point that could feed a termination decision to see if it will trigger the new law. This could include productivity scoring, attendance tracking, route optimization, quality monitoring, customer review aggregation, and performance dashboards. Ask your vendors directly whether their product produces a score, classification, or recommendation.

2. Define what “primarily relied upon” means for your organization. The statute does not define it, so you need an internal standard and consistent application. Decide now which tools you will treat as triggering the law and which you will treat as background information, and document the reasoning.

3. Build a human review protocol with a paper trail. Designate who reviews, what evidence they must examine, and how they record their conclusion. A checkbox confirming the reviewer “agreed” may not survive the burden-shifting structure of the new law. Capture what independent evidence the reviewer examined and what they concluded about the system’s accuracy.

4. Draft your post-use notice template now. Build it into your termination packet so it is delivered at the moment of the decision, not days later. Prepare translated versions for every language you use for routine communications, and identify the human contact by role rather than by name so the template survives turnover.

5. Train your managers and build a new process. Frontline supervisors need to treat a flag from the system as a reason to look more closely at the situation. Build a workflow for responding to annual employee data requests that produces a meaningful description while anonymizing information about customers and other employees.

Conclusion

We will monitor developments related to SB 947 and provide updates as warranted, so make sure you are subscribed to Fisher Phillips’ Insight System to get the most up-to-date information directly to your inbox. If you have questions, contact your Fisher Phillips attorney, the authors of this Insight, or any attorney in our California offices or our AI, Data, and Analytics Practice Group.