Imagine spending hours tailoring your resume for a job, only to be rejected before an actual human being sees and screens your application. This situation is no longer hypothetical – it is the reality for those in the job market.
Today, most people who apply for a job in California will be evaluated, at least in part, by software. Many employers use artificial intelligence and other automated tools to screen resumes and look for keywords, administer personality assessments to determine if a candidate is a good fit for the company culture, and even analyze video interviews sent beforehand. Gone are the days when you have to prepare for an initial interview with a company representative to shoot your shot. Usually no one tells the applicant that all of this software screening has happened, and the rejection email reads the same either way.
Since October 1, 2025, California has had regulations aimed squarely at that situation. The rules do not prohibit employers from using AI in recruitment, hiring, and promotion. However, the California Civil Rights Council clarified that one of the aims of the law is to ensure that employers cannot avoid liability under the Fair Employment and Housing Act (“FEHA”) just because an automated system made a decision and not a human being. What the rules do is close the gap an employer might otherwise try to open between “we decided” and “the system decided.”
The governing rule
The Civil Rights Council writes the regulations interpreting California's main workplace anti-discrimination law, the Fair Employment and Housing Act. In 2025, it adopted rules covering what it calls an automated-decision system, defined as:
“A computational process that makes a decision or facilitates human decision making regarding an employment benefit.”
Two phrases carry most of the weight.
Facilitates human decision making means the rules reach tools that merely advise. A hiring manager who narrows a stack of applicants using scores the software produced is still inside the regulation, even though an actual person made the final call.
Employment benefit is defined in the same regulations to cover hiring, promotion, compensation, selection for training programs that lead to promotion, discharge, and other terms, conditions, and privileges of employment. The rules follow the employment relationship past the application stage.
At its core, the regulation is straightforward: an employer cannot use an automated-decision system in a way that discriminates against an applicant or employee based on a protected characteristic under FEHA. Whether the employment decision came from a human being or an AI tool, if the result is unlawful discrimination, FEHA governs. Race, sex, age, disability, national origin, religion, and the other protected characteristics apply exactly as they would if a person had made the same call.
The regulations also name the specific tools they have in mind. An automated-decision system includes software that makes predictive assessments about an applicant; measures skills, dexterity, or reaction time; measures personality traits, aptitude, attitude, or “cultural fit”; screens, evaluates, categorizes, or recommends candidates; directs job advertisements to targeted groups; screens resumes for particular terms or patterns; analyzes facial expression, word choice, or voice in online interviews; or analyzes applicant data bought from third parties.
Ordinary business software is carved out. Word processors, spreadsheets, calculators, databases, and similar tools fall outside the definition, so long as they are not the thing making the decision about an employment benefit.
What this changes in practice
Employers cannot shift the blame to their AI vendor. One of the important aspects of the regulation is that employers cannot avoid liability by pointing the finger at the software vendor. Most employers do not build these AI tools themselves. They purchase or license them from third-party vendors. However, the distinction makes no difference when it comes to responsibility. The regulations define an “agent” as any person acting on the employer's behalf to perform a traditional employer function, including recruiting, screening, hiring, promotion, and decisions about pay, benefits, or leave, and including when those activities are carried out through an automated-decision system. The regulations then state that an agent of an employer is also an employer for purposes of the Act. This tracks the California Supreme Court's 2023 decision in Raines v. U.S. Healthworks Medical Group, which held that a business-entity agent performing FEHA-regulated functions on an employer's behalf can itself be liable under FEHA. We wrote about Raines when it came down, and the reasoning is what makes “the vendor's algorithm did it” a weak answer.
AI records now must be kept for four years. The recordkeeping regulation previously required employers to preserve personnel and employment records for two years. It now requires four, running from the date the record was made or the date of the personnel action, whichever is later. The amended list of what must be preserved was expanded to name automated-decision system data explicitly, alongside applications, personnel records, and selection criteria. Automated-decision system data is defined broadly enough to include the data fed into the tool, the data it produced, and the data used to build or customize it for that employer.
For an employee, the practical effect is about proof. Discrimination that runs through software leaves a trail in the software, and that trail is now required to exist for twice as long as before, making it easier to investigate and prove discrimination claims.
Whether the employer tested the tool is evidence. The regulations place emphasis on whether an employer evaluated its AI tool for bias before using it. It states that evidence, or the lack of evidence, of anti-bias testing or similar proactive efforts is relevant both to a discrimination claim and to an employer's defense. They go further and list what counts: the quality, efficacy, recency, and scope of the effort, the results, and the employer's response to those results. If in the course of testing and screening the tool, an employer found a disparity and discovered that the AI disproportionately excluded applicants from a protected class but continued using the system without addressing the issue, it could be significant evidence in a later FEHA lawsuit.
Two familiar limits now follow the software. California already barred employers from asking about criminal history before a conditional offer. The amended regulations specify that the prohibition covers inquiries made through an automated-decision system, so a background-screening tool run at the application stage is treated the same as the question on the form. The same principle applies to disability-related inquiries. The regulations explain that AI-driven assessments including personality tests, questions, puzzles, games, or other challenges likely to elicit medical information about a disability may violate the rules and may constitute an unlawful medical inquiry.
What the rules do not do
Being clear about the limits matters as much as describing the rule.
Applicants do not have a general right to notice. California came close to requiring notice. Senate Bill 7, the proposal known as the “No Robo Bosses Act,” would have required employers to notify workers before using automated decision systems affecting them and would have barred relying on those systems alone for discipline or termination. It was vetoed on October 13, 2025, and the veto was sustained on March 2, 2026. So, an applicant generally has no standalone statutory right to demand notice that software screened them.
The rules do not create a new claim. These are regulations interpreting an existing statute. They clarify that FEHA reaches automated decisions and sharpen what evidence matters. An employee still has to establish discrimination under the standards that already applied, and employers retain the defenses that already existed. The regulations say so directly: the prohibition operates “subject to any available defense.”
Small employers are outside FEHA's core coverage. The definition of employer reaches businesses regularly employing five or more individuals. Different rules and thresholds apply elsewhere in California law, and harassment provisions reach more broadly.
Using AI is not itself unlawful. Nothing here makes it improper for an employer to use these tools. The question is always whether a particular tool, as actually used, produced discrimination on a protected basis.
What this means if you are looking for work in California
- Software that screens, scores, or ranks you is covered by the same anti-discrimination rules that apply to a human manager. A decision does not become lawful because a system produced it.
- A rejection generated by an algorithm is not automatically a violation. What matters is whether the tool disadvantaged you because of a protected characteristic.
- Assessments that look like games can still be inquiries. If a timed challenge or personality test seems built to surface a medical or psychological condition, that is worth paying attention to, particularly before any offer is made.
- Questions about criminal history are still off-limits before a conditional offer, including when a screening tool asks them.
- Keep your own record. Save the posting, the assessments you were asked to complete, the names of the platforms, and the correspondence. The employer's four-year retention obligation is not a substitute for your own contemporaneous copies.
- Timing rules still apply. Deadlines for filing with the Civil Rights Department and for bringing suit are unchanged by these regulations, and they are shorter than most people expect.
If you believe an employment decision about you was influenced by a protected characteristic, whether a person or a system made the call, an employment attorney can evaluate the specific facts. Java & Jebreil represents employees across California in discrimination and harassment, retaliation, and wrongful termination matters.
Sources
- California Civil Rights Council – Employment Regulations Regarding Automated-Decision Systems (rulemaking record; effective October 1, 2025)
- Final text of the automated-decision-system employment regulations
- California Legislative Information – Senate Bill 7 status and veto history
This article is general information about California law and is not legal advice. It does not create an attorney-client relationship, and it does not address the facts of any particular situation. Laws and regulations change, and how they apply depends on the circumstances. For advice about your own situation, consult a licensed attorney.
