California Tightens Scrutiny of AI Hiring Tools Amid Reports of Racial Bias

California Tightens Scrutiny of AI Hiring Tools Amid Reports of Racial Bias

California employers that use artificial intelligence (AI) to screen job candidates face an increasingly complex compliance landscape as state regulators clarify that existing anti-discrimination laws apply fully to automated decision systems (ADS). At the same time, lawmakers are advancing legislation imposing new transparency, testing and governance requirements on both employers and ADS developers.

The emerging framework reflects a broader regulatory message: employers cannot avoid liability for discriminatory hiring decisions simply because those decisions were made or heavily influenced by AI. Instead, organizations remain responsible for ensuring automated systems comply with longstanding employment discrimination laws while preparing for additional AI-specific obligations.

The heightened scrutiny comes as new research from Stanford University’s Institute for Human-Centered Artificial Intelligence raises fresh concerns about the performance of widely used hiring algorithms. The study examined more than four million job applications submitted to 150 employers and found significant racial disparities in recommendations generated by a leading third-party hiring platform, with Black and Asian applicants disproportionately disadvantaged. Researchers estimated that approximately 40,000 additional applications from qualified Black and Asian candidates would have advanced had the system recommended all racial groups at equal rates.

According to an analysis by law firm Hanson Bridgett, California’s Civil Rights Department (CRD) clarified in regulations that took effect in October that employers with five or more employees remain subject to the California Fair Employment and Housing Act (FEHA) when using ADS in employment decisions. The rules cover automated assessments including computer-based tests, skills evaluations, puzzle games and AI-powered video interview analysis.

Perhaps most significantly for human resources and compliance executives, the regulations make clear that existing anti-discrimination and disability accommodation requirements continue to apply regardless of the role AI plays in a hiring decision. If an applicant’s disability affects performance on an automated assessment, employers must provide reasonable accommodations consistent with FEHA. Companies therefore should ensure either their AI vendors provide accommodation request mechanisms or establish their own process for applicants seeking accommodations.

The CRD regulations also expand employers’ recordkeeping responsibilities. Companies must retain ADS-related employment records, including system inputs, outputs and data used to design or customize the technology, for at least four years to facilitate regulatory oversight and potential discrimination investigations.

Existing California hiring restrictions likewise extend to automated systems. Under the state’s Fair Chance Act, employers may not inquire into or consider an applicant’s criminal history before extending a conditional offer of employment. Hanson Bridgett noted that employers should audit AI screening tools to verify criminal history information is neither accessed nor incorporated into recommendations before that stage, emphasizing that reliance on an ADS would not shield employers from Fair Chance Act liability.

California lawmakers are also considering legislation that would substantially expand AI-specific compliance obligations, per Hanson.

The proposed Automated Decisions Safety Act (AB 1018) would require employers and AI developers to conduct regular bias testing, notify applicants whenever AI is used in employment decisions, and explain the basis for automated decisions upon request. The measure would also impose direct legal obligations on AI vendors, not just employers, representing a significant expansion of accountability across the AI supply chain.

A separate measure, AB 1898, would require employers to provide advance written notice whenever an automated decision system affects an employee or applicant, obtain signed acknowledgments confirming receipt of those notices, and maintain an annual inventory of all AI tools used in employment decisions.

The Hanson Bridgett analysis also highlighted separate obligations under California Consumer Privacy Act regulations. Beginning Jan. 1, employers must complete formal risk assessments before using automated decision-making technology for significant hiring decisions. By April 1, 2028, organizations must submit summary reports to the state, including executive certification that required assessments have been completed.

The Stanford research underscores why regulators are moving aggressively. Beyond identifying disparate impacts against Black and Asian applicants, researchers found evidence that reliance on the same third-party hiring platforms across multiple employers can create what they describe as “systemic rejection,” with applicants repeatedly rejected across employers because they are evaluated by the same underlying algorithm. The researchers argued that independent evaluation of hiring algorithms will be essential for evidence-based AI governance as employers increasingly rely on automated hiring technologies.