California's 2025 AI Employment Rules Are Here
California AI employment rules are about to change how hiring and promotions work across the state. Starting October 1, 2025, the state’s civil rights regulations expressly apply FEHA’s discrimination protections to Automated-Decision Systems (ADS)—tools that help make or facilitate employment decisions (think resume screeners, online assessments, targeted job ads, or algorithmic promotion engines). If an automated system produces biased outcomes, it’s the employer’s problem—and potentially their vendor’s too.
For employees, this means new leverage: better records, clearer definitions, and explicit recognition that AI-infused processes can violate FEHA just like any human-run process. If you think an algorithm cost you a job, promotion, accommodation, or fair treatment, you now have specific hooks to demand testing, data, and explanations under the California AI employment rules.
What Exactly Counts as an “Automated-Decision System” (ADS)?
Under the regulations, an ADS is a computational process that makes a decision or facilitates human decision-makingabout an employment benefit (e.g., hiring, training, promotion). Concrete examples include:
Computer-based tests that screen, evaluate, categorize, or recommend applicants or employees, including tools that measure traits like skill, reaction time, or “cultural fit.”
Targeted job ads or recruiting that aim messages at particular groups.
Resume screeners that look for certain terms or patterns.
Analytics on interview video or audio (voice, facial expression, word choice).
Data pulled from third parties to rate or sort applicants.
The definition excludes general office software (word processors, spreadsheets, email, firewalls, etc.) so long as it doesn’t make decisions about an employment benefit.
Who’s Covered—and Why “Agent” Status Matters
The rules cover employers, employment agencies, and labor organizations, plus any “agent” acting on an employer’s behalf—including when activities are conducted in whole or in part through an ADS. In plain English: if a vendor screens applicants for you using AI, that vendor can be treated as an agent (and even as an “employer” under FEHA in some contexts). This aligns with the California Supreme Court’s decision in Raines v. U.S. HealthWorks, which recognized direct FEHA liability for certain business-entity agents.
Why workers should care: When your case involves a third-party platform (ATS, testing vendor, background-screening service), you can seek discovery from both the employer and the vendor, and you may have a direct claim against the vendor depending on the facts.
Do Employers Have to Do “Bias Testing”? (Short Answer: No—but it’s Crucial Evidence)
The regulations do not mandate bias testing. However, in any discrimination claim involving an ADS, the presence or absence of anti-bias testing or similar proactive efforts—and the quality, efficacy, recency, scope, and results—is explicitly relevant. That means if an employer never tested its tool, tested it years ago on the wrong population, or ignored bad results, that can undermine their defense. Conversely, solid, recent testing may help them.
Employee tip: Ask for all testing plans, audit reports, validation studies, and results. If those don’t exist or are stale, that absence can support your theory of discrimination under the California AI employment rules.
Recordkeeping: Four Years of Ammunition for Workers
A major win for transparency: employers must preserve records for four years, including “automated-decision system data” (data used in or resulting from an ADS, and data used to develop or customize it). The four-year duty also covers applications, personnel records, selection criteria, and referral records—in short, the very materials you’ll need to prove an AI-driven pattern or practice.
Employee tip: In a legal hold or pre-litigation letter, identify ADS data by name and demand preservation (e.g., model inputs/outputs, scores, rankings, feature importances, thresholds, error logs, change logs, and vendor deliverables).
The Legal Standards Still Apply—Now with AI in the Frame
The regulations weave ADS into existing FEHA doctrines:
Disparate treatment and adverse impact: Using an ADS or selection criteria that discriminates on a protected basis is unlawful unless the employer proves a recognized defense (e.g., business necessity, with no less-discriminatory alternative available).
Across protected classes: The text cross-references ADS usage in sections covering sex/pregnancy, disability, national origin/ancestry, and age—making clear that automated tools are judged under the same FEHA protections.
Criminal history & “Ban the Box”: Employers can’t use ADS to probe criminal history before a conditional offer or to make decisions inconsistent with FEHA’s Fair Chance rules.
What this means for a plaintiff case: You still run the classic FEHA playbook—but now you connect the dots between algorithmic steps and protected-class impacts, using the employer’s own ADS data to establish adverse impact and to test their claims of business necessity.
Practical Red Flags Workers Should Watch For
Automated rejections with no explanation (e.g., “system determined you’re not a fit”). Ask how the score was generated and what inputs were used.
Ungrounded “fit,” “culture,” or “personality” screens that correlate with protected traits or disabilities.
Targeted job ads that never reach certain age groups, women, people with disabilities, or specific national origins.
Video or voice analysis of candidates that rates facial expression or speech patterns. Demand evidence of accessibility accommodations and validation.
No opt-out or accommodation path for applicants with disabilities when tests require particular interfaces or speeds.
How to Document and Build Your Case (Step-by-Step)
Save everything: job post links, screenshots, auto-emails, score displays, rejection times, and application portal messages.
Request your data: Ask the employer to provide your ADS-related records—scores, rankings, explanations—citing the California AI employment rules and FEHA.
Write to HR (in California-aware terms): Ask what system was used, its purpose, whether it was validated or bias-tested, what features it considered, and whether alternatives were available.
Note protected facts: If you’re pregnant, returning from leave, over 40, disabled, or a member of a protected class, record any timing or statements that connect your status to the decision.
If you need accommodation: Request one in writing (e.g., extra time, accessible format, non-video option). Employers must consider accommodations for tests—including ADS-administered tests.
Talk to a lawyer early: Early preservation demands can secure four years’ worth of ADS data—gold for proving patterns.
What We’ll Ask For in Discovery
Expect a targeted discovery plan aimed at both the employer and any agent/vendor:
System identity & purpose: Name/version, vendor, modules, change logs, decision thresholds.
ADS data: Inputs, outputs, scores, rankings, audit logs, error logs, fairness metrics. (The regs define “automated-decision system data” and require 4-year preservation.)
Bias testing & validation: Protocols, statistical methods, recency, scope, efficacy, results, remediation steps.
Business necessity & alternatives: Why this tool, what goals it serves, and whether less discriminatory alternatives were considered or tried (key to adverse-impact analysis).
Accommodation handling: Policies and records showing how applicants with disabilities could access equivalent, validated assessments.
FAQs
Are California employers required to audit or bias-test their AI?
No. But the existence, quality, and recency of anti-bias testing (or the lack of it) is relevant evidence in a discrimination case. That’s a powerful lever for plaintiffs.
Can I sue the vendor that built or ran the AI?
Depending on facts, yes—vendors can be treated as an “agent” (and thus potentially an “employer”) under FEHA. Rainesconfirms business-entity agents can face direct FEHA liability in appropriate circumstances.
What if the company says, “A human made the final call”?
Because the definition covers systems that facilitate human decision‑making, FEHA applies even when a person “makes the final call” after using an ADS to screen, rank, or shape the pool.
I was auto-rejected before I could explain a disability or request accommodation. Is that legal?
Not if the process discriminates or fails to reasonably accommodate. Employers must ensure tests (including those administered via ADS) fairly measure job skills and provide appropriate accommodations.
What records should I ask for first?
Start with all ADS data related to your application: inputs, outputs, scores, rankings, adverse-impact analyses, and any bias testing results. The employer must keep such records for four years.
Bottom Line for California Employees
The California AI employment rules don’t reinvent FEHA—they plug AI directly into FEHA’s existing protectionsand give you stronger tools to prove discrimination: broader definitions, clear agent liability, four-year recordkeeping, and explicit recognition that the quality and recency of bias testing matter. If you suspect an algorithmic tool hurt your chances, act quickly to preserve data and build your case.
Talk to a Lawyer Who Understands AI Cases
If you think an AI‑driven decision harmed you, we can help. At H&A Law Office, PC, we represent employees across California in discrimination, retaliation, and wrongful termination matters—now including AI‑related claims under FEHA.
Call: (424) 322‑2482
We’ll evaluate your situation, send preservation demands, and push for the ADS data, testing records, and vendor documents that can prove your case.
This post is intended for informational purposes only and does not constitute legal advice. Laws and regulations are subject to change, and their application can vary based on specific circumstances. For personalized legal assistance, please reach out to H&A Law Office, PC.
