Audit the system by tracing how it affects employment decisions—not by relying on the vendor’s label or the presence of a human reviewer. Inventory tools used in recruiting, screening, assessment, hiring, promotion, discipline and termination; map who controls each decision; examine job relevance, selection outcomes, accessibility and accommodation; and preserve the records California requires. California’s automated-decision employment regulations took effect October 1, 2025. A separate 2026 announcement describes a restriction on relying only on AI for discipline or termination, but its precise requirements should be checked against the enacted bill text.
Which systems belong in the audit?
Start with function and effect. California’s Civil Rights Department (CRD) describes an automated-decision system as a computational process that makes a decision or facilitates human decision-making about an employment benefit. The definition can reach systems using AI, machine learning, algorithms, statistics or other data-processing techniques. A tool does not fall outside the audit simply because a person formally makes the final call.
As an Amazon Associate I earn from qualifying purchases.
Examples in the regulations include computerized tests and assessments, applicant screening and recommendations, résumé keyword or pattern screening, targeted recruiting materials, analysis of facial expressions, word choice or voice in online interviews, and analysis of third-party applicant or employee data. General-purpose technical tools are excluded when they do not make an employment-benefit decision.
Include the full employment decision path
- Recruiting: sourcing, job-ad targeting and delivery, and tools that influence who sees or responds to an opportunity.
- Applicant processing: résumé parsers, filters, rankings, recommendations and eligibility screens.
- Assessment and interviewing: tests, puzzle games, video or voice analysis, and tools that interpret behavior or responses.
- Workforce decisions: systems used to evaluate, rank or recommend employees for opportunities or actions, including promotion, discipline or termination.
Record each system’s purpose, version, vendor, California use, affected roles, inputs, outputs and the decision points it touches. Include systems that screen people out, direct a decision-maker’s attention or shape the pool of candidates—even if the tool is described as advisory.
#1 Best Overall
How should you map responsibility and decision flow?
Document what happens from input to consequence. Identify the employer owner, vendor and any other agent involved in recruitment, screening or hiring; the human reviewer; who can override a recommendation; and what actually happens when a person receives a particular score or ranking. Employers’ agents may act directly or indirectly in these activities, including through an automated system.
For each step, distinguish between what the system technically produces and how the organization uses it. A recommendation may function as a decisive screen if staff routinely follow it or if lower-ranked applicants are never reviewed. Record override authority and actual human actions rather than assuming that a nominal human checkpoint changes the system’s effect.
How do you test job relevance and discrimination risk?
Review the selection practice and its outcomes, not just the model’s stated purpose. California’s regulations address selection policies or practices that have an adverse impact on protected groups. They describe adverse impact as including substantial disparities in selection rates and permit evidence beyond statistics. The sources do not establish a single AI-specific test, metric or threshold that acts as a safe harbor.
Connect inputs and criteria to the job
- Write down the job requirements and the criteria the system uses to score, rank or screen candidates.
- Check whether each input measures a skill or qualification relevant to the work, rather than a proxy unrelated to essential duties.
- Inspect data sources and features for proxies that could disadvantage protected groups, even when a protected trait is not expressly supplied to the system.
- Ask whether the criteria reflect the actual role and are applied consistently across candidates.
Compare outcomes across stages
Examine selection and adverse-action rates at relevant stages, such as application review, assessment, interview and advancement. Interpret differences in light of the job, applicant pool and selection criteria; preserve that context along with the analysis. If a disparity appears, investigate what feature, threshold or stage may contribute to it and document the organization’s assessment and response. Do not present one statistical method or cutoff as a California AI compliance guarantee.
CRD has illustrated the risk with a hiring tool that reproduces patterns from a male-dominated workforce and with ad delivery that reinforces gender or racial stereotypes. Those examples illustrate possible mechanisms; they are not findings about a particular vendor or evidence of a measured industry-wide rate.
How should you test accessibility, accommodations and inquiries?
Accessibility and accommodation are core audit tracks. CRD warns that ability tests, dexterity or reaction-time measures, and analysis of tone, facial expression or behavior may disadvantage people with disabilities or other protected characteristics. A system’s use does not substitute for an individualized assessment.
Rank #3
Test the candidate or employee experience
- Try the application and assessment with relevant assistive technology and alternative formats.
- Check whether instructions, time limits, input methods and interview formats create barriers unrelated to the job skill being assessed.
- Confirm that people can request a reasonable accommodation through a usable channel and that requests reach someone with authority to respond.
- Determine whether an alternative selection method can serve the employer’s goal with less discriminatory impact.
- Document accommodation handling and individualized review without exposing sensitive information more widely than necessary.
Review questions and inferences
Inspect both the questions a system asks and what it infers from a person’s answers, voice, face or behavior. CRD says assessments—including tests, questions and puzzle games—that elicit disability information may constitute unlawful medical inquiries. The regulations address pre-offer medical or psychological inquiries, including ones administered through automated systems. Review interview analysis and third-party data for inferences that may create the same concern.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWhat records should the audit preserve, and for how long?
CRD says employers and covered entities must keep employment records, including automated-decision data, for at least four years. The final regulation text also specifies four-year preservation for certain records and applicant-identification information. Which records and triggering dates apply can depend on the organization’s circumstances, so have counsel confirm the retention schedule.
A useful audit file can include the system name and version, configuration, vendor documentation, data sources, job criteria, outputs, decision records, human actions and overrides, selection-rate analyses, accommodation handling, complaints and corrective actions. Use access controls for sensitive information, and make sure records can be exported and retrieved if the organization needs to investigate a decision or assess a change.
How do the California requirements fit together?
FEHA automated-decision employment regulations
The Civil Rights Council’s final employment regulations were approved by the Office of Administrative Law on June 27, 2025, and took effect October 1, 2025, according to CRD. They provide the central framework for assessing systems that make or facilitate decisions about employment benefits, including the selection-practice, adverse-impact, accommodation and recordkeeping issues described above.
Privacy rules
The California Privacy Protection Agency’s laws page lists CCPA regulations effective January 1, 2026 and an adopted package addressing automated decisionmaking technology, risk assessments and cybersecurity audits. That status information does not mean every employer or employment system is covered. Assess applicability based on the organization’s status, the information it collects, the processing involved and relevant exceptions; consult the operative rules for the specific obligation.
Recommended Free Tools
State-agency inventory process
California’s Department of Technology describes a high-risk automated decision system inventory and reporting process for state agencies. Its FAQ gives a generative AI résumé-ranking system that influences whether a public employee advances or is rejected as an example. This is a distinct public-agency process, not a general filing duty for every private employer.
Best Value
2026 announcement concerning discipline and termination
On September 30, 2026, the Governor’s announcement about SB 947 described a worker-protection measure that prohibits employers from relying only on AI when making disciplinary or termination decisions. The announcement does not establish the statute’s exact definitions, covered employers, exceptions, remedies or effective date. Check the enacted, chaptered bill text before turning the announcement into a precise legal requirement or operational deadline.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical audit workflow
- Inventory systems and decision points. List tools used in recruiting, screening, assessment, interview analysis, ranking, recommendations and workforce decisions. Record each system’s purpose, version, affected roles and California use.
- Map responsibility and consequences. Identify the employer owner, vendor, agents, reviewers, override authority and downstream effects. Note whether a system assists a decision or effectively determines who advances, is hired, promoted, disciplined or terminated.
- Validate criteria. Tie inputs and scoring to documented job requirements. Examine proxies and criteria that may exclude protected groups or are unrelated to essential duties.
- Analyze outcomes. Compare selection and adverse-action rates across relevant protected groups and stages. Investigate disparities, record data limitations and context, and assess any asserted job-related and legally sufficient defense.
- Test accessibility and accommodation. Check assistive-technology compatibility, alternative formats, accommodation channels, job-skill relevance and individualized review.
- Inspect recruiting and inquiries. Review ad targeting and delivery, application questions, interview analysis and third-party data inferences for discriminatory effects, discouragement or improper medical inquiry.
- Preserve evidence and remediate. Retain covered records for the applicable period. Record versions, tests, decisions, explanations, human overrides and corrective actions; escalate material risks to counsel and the responsible decision-maker.
- Check parallel regimes. Determine whether California privacy rules apply; if the organization is a state agency, separately assess the high-risk ADS inventory process. Verify SB 947 against its enacted text before relying on details not established by the Governor’s announcement.
How to compare systems or vendors
There is no state-approved product rating scheme established by these materials. Use the same evidence-based questions for each system under consideration:
Quick Recap
- What decisions does it influence, and how automated is the process in practice?
- Are its criteria job-related and documented?
- Can the employer examine outcomes and investigate disparities?
- Does it support accessible use and accommodation workflows?
- What data sources and potential proxies does it use, and can records be exported?
- Will the vendor explain system behavior and cooperate with an audit?
- Is there meaningful human review, override authority and a path to raise concerns?
- Can the organization preserve records and investigate incidents over time?
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
