The Civil Rights Council’s automated decision system rules reach any California employer with five or more employees, so they cover a company of 10 to 150 people. The privacy agency’s rules start January 1, 2027 and reach only businesses above the CCPA thresholds, so many small employers fall outside them and some do not.
The Civil Rights Council’s automated decision system rules reach any California employer with five or more employees, so they cover a company of 10 to 150 people. The privacy agency’s rules start January 1, 2027 and reach only businesses above the CCPA thresholds, which many small employers stay under and some do not. Check both this month.
I am Mac Malbrough. I hold a J.D. and an MS-HRM, I am not an attorney, and I write for California employers who use an applicant tracking system and wonder which AI rules apply to them. Where a point is my reading and not the regulation’s words, I say so.
Which California AI hiring rules reach a small employer?
Two sets of rules matter, and they have different reach. The Civil Rights Council regulations under the Fair Employment and Housing Act apply to you. The California Privacy Protection Agency’s rules apply to you only if your business crosses a privacy-law threshold.
The Civil Rights Council rules took effect on October 1, 2025, per the Civil Rights Department announcement and the Council’s rulemaking page. The final text defines an employer as one regularly employing five or more individuals. I found no higher threshold for the automated decision rules, so every employer with 10 to 150 people is inside them.
The privacy rules are different. They start from the California Consumer Privacy Act, which covers employee and applicant data and reaches only businesses that meet one of three tests. I explain those in section three. Legislation is a third track: the Governor vetoed SB 7 on October 13, 2025, per his veto message, and my guide to the No Robo Bosses Act covers the bill that followed.
What do the Civil Rights Council rules require of you?
They require that your tools do not discriminate, and that you keep the records. The rules do not ban AI. They apply the existing anti-discrimination law to it.
The definition is broad. Under section 11008.1(a), an automated decision system is a computational process that makes a decision, or helps a human make one, about an employment benefit. The listed examples include screening resumes for particular terms or patterns. Word processors, spreadsheets and spam filters are excluded when they do not make a decision about an employment benefit. In my reading, an applicant tracking system that ranks, scores or filters candidates fits the definition. My guide to how applicant tracking systems screen candidates explains what these tools do.
Four duties follow from the final text.
You may not use an automated decision system or selection criteria that discriminate on a basis the Fair Employment and Housing Act protects, subject to any available defense (section 11009(f)).
Evidence of anti-bias testing, or the lack of it, can be relevant to a claim or a defense, including the quality, recency and scope of the testing and your response to the results. It is not a safe harbor.
An agent of an employer is also an employer, and the definition reaches people who screen applicants for you, including through an automated system (sections 11008(a) and (e)(3)). I read that to make a screening vendor a possible co-employer under the law, although the text never says vendor.
You must keep applications and automated decision system data for four years, up from two (section 11013(c)), measured from the record or the personnel action, whichever is later.
Two older rules now reach your software as well. The criminal history rule bars inquiries before a conditional offer, and section 11017.1(a)(1) lists inquiries made through an automated decision system. The medical rule bars examinations or inquiries before an offer, and section 11071(e) includes any made through an automated system, such as a game that is likely to elicit information about a disability.
When do the privacy agency’s rules apply to you?
They apply when your business meets a CCPA threshold and uses automated decisionmaking technology for a significant decision about a person. That is a smaller group than the group the Civil Rights Council rules reach.
The agency’s regulations were approved by the Office of Administrative Law on September 22, 2025, per its Notice of Approval. The approved text requires a business already using the technology for a significant decision to comply by January 1, 2027. A significant decision includes hiring, assigning work, pay, promotion, demotion, suspension and termination.
Your business is covered under Civil Code 1798.140(d) if it does business in California and meets any one of three tests: annual gross revenue above the adjusted threshold, buying, selling or sharing the personal information of 100,000 or more consumers or households, or earning half its revenue from selling or sharing personal information. The agency’s CPI adjustment page lists the revenue figure as $26,625,000 and says adjustments come every odd-numbered year. The agency’s FAQ confirms the figure and says employees and applicants count as consumers.
Many employers with 10 to 150 people stay under every test, but do not assume it. A 150-person firm can pass $26.6 million in revenue. Check your last full year of gross revenue, and ask whether you sell or share customer data.
One more point helps. The rules define the technology as something that replaces or substantially replaces human decisionmaking, and they say a person counts only if they know how to interpret the output, review it and other information, and have the authority to change the decision. A tool whose output a trained person reviews with real authority may fall outside the definition, which is my reading of the text.
What should you ask your applicant tracking vendor this month?
Ask in writing, and save the answers. A vendor that screens for you is part of your compliance, and its answers become part of your record. I use seven questions.
Does the product rank, score, filter or reject applicants, or only store them?
What criteria does it use, and can you see and change them?
What anti-bias testing has been done, when, by whom and with what results?
Can you export applications and automated decision data for four years?
Does it ask about criminal history or medical conditions, and at what stage?
Does it use games or assessments that could elicit disability information?
Does it post jobs to boards, and does it carry your pay scale?
The last question comes from Labor Code 432.3, which requires employers with 15 or more employees to include a pay scale in postings and to give the pay scale to a third party that publishes the posting. If your vendor auto-posts, check what it actually publishes. My guide to HR software for talent acquisition covers the selection side, and my AI and HR page collects the rest of my AI work.
What should a person review before a tool rejects anyone?
A person who can say yes, who knows what the output means and who has looked at more than the score. That is both good practice and the privacy rules’ definition of human involvement.
Build the checkpoint on three tests. The reviewer understands how the tool reached its output, reviews the output and the application together, and has authority to overrule it. A recruiter who clicks approve on every rejection does not meet that test. Record who reviewed each rejection and when, and keep it with the application for the four-year retention period.
Start the review with the knockout rules. Look at every automatic rejection your system applies: minimum experience, gaps in work history, location, a degree requirement and any question about convictions. A rule that screens out applicants at higher rates because of a protected trait is the kind of problem the regulations address.
McKinley holds a J.D. but is not a licensed attorney. Articles here are general information, not legal advice. For your specific situation, talk with an employment attorney.
Your best people leave because the next title does not exist. A career lattice offers growth sideways and deeper, not only up. Three skill levels per role, planned lateral moves, a pay…
Your people want to be seen and you are too busy to notice out loud. A ten-minute weekly routine, four tests for praise that lands, three traps, and the pay and tax…
If one person holds payroll, benefits renewals and the I-9s in their head, a resignation or a sick week becomes a legal deadline. A five-column task inventory, the three-week test, a documentation…