AI Made Me Do It: What Employers Should Know About AI Employment Decisions

Artificial intelligence (AI) tools have become ubiquitous in employment decision-making, from résumé screening to shift scheduling and workforce reduction targeting. While AI promises efficiency and data-driven objectivity, it also creates significant legal exposure under anti-discrimination laws, testing standards, and employment regulations.  Importantly, in the context of AI and employment decisions, the employer’s intent does not matter; employers cannot argue that because AI did it, they are immune from liability. Below we discuss some practical guidance for employers based on new regulations and recent case law.

New California AI Regulations

Recently, in California, the Civil Rights Council issued updated AI regulations applicable to employment decisions under FEHA. Notably, these regulations apply to all employers who use “Automated Decision Making” to facilitate human decision-making with respect to the recruitment, hiring, and promotion of job applicants or employees. The regulations broadly define Automated Decision Making (ADS) as any “computational process that makes a decision or facilitates human decision making regarding an employment benefit” that “may be derived from and/or use artificial intelligence, machine-learning, algorithms, statistics, and/or other data processing techniques.” 2 Cal. Code Regs. § 11008.1(a).

The regulations prohibit employers from using ADS to discriminate against applicants or employees based on protected categories, and any facially neutral ADS selection tools that have an “adverse impact” on applicants or employees based on a protected characteristic are impermissible under FEHA, unless the employer or covered entity can show that the selection practice is “job-related and consistent with business necessity.” The regulations also cover “proxies” for protected characteristics (i.e. Zip Codes or neighborhoods; Commute distance; Facial expression Tone of voice or speech patterns; Graduation year; Employment gaps.) While California is one of the first states to begin to address AI and employment decisions, many other states are following suit, so employers in all jurisdictions should be aware of the risk.

Put simply, employers may be exposed to potential liability for what their AI tools are doing, or, are not doing, especially as it relates to employees that are part of a protected class.

Recent Cases Provide Cautionary Tales For Employers

In July, 26 current and former employees filed suit against Meta alleging Meta used internal artificial intelligence systems to select staff for layoffs without neutralizing metrics for protected medical or parental leave, effectively penalizing workers for taking legally permitted absences.

The Complaint alleged Meta used “keystroke-and activity-monitoring data” and “algorithmically assisted performance ranking” to rank, and select employees for its reduction in force. Plaintiffs argue that the inputs the artificial intelligence used, improperly left out employees with disabilities and on protected leaves and resulted in unlawful terminations..

The lawsuit poses several critical questions, including:

• Whether Meta used employee accommodation or leave status, or proxies for protected activity as a negative factor, and if not, how it accounted for these protected activities;
• Whether Meta tested the process for disparate impact prior to rolling out the process; and
• Whether the reduction in force process resulted in an unjustified disparate impact.    

In August 2024, the US Court of Appeal for the Seventh Circuit ruled against Wal-Mart in a case alleging that Wal-Mart’s AI systems illegally discriminated against an employee with down syndrome. Plaintiff lost her job with Wal-Mart after the company changed its work scheduling policies and she had difficulty working the new shift she was assigned due to her disability. Walmart issued a directive that managers were to cease making manual adjustments to AI-generated staff work schedules in the absence of a business justification for doing so. As a result, and despite Plaintiff’s complaints, Wal-Mart began issuing attendance infractions, and ultimately terminated her employment.

What Should Employers Do? 

With that in mind, employers should take the following steps with AI:

• Analyze AI systems that are used for any employment decision (hiring, scheduling, performance evaluation, pay, leaves, accommodations, restructuring, firing, etc.), and ensure these tools account for disparate impact on employees in protected categories and those who engage in protected activities.
• Frequently test for bias in AI tools. Relying on blanket assertions that the tool is "objective" or "unbiased" without empirical support is not a sufficient test.
• Evaluate whether AI treats protected leave or accommodations as underperformance, such that it creates a negative performance signal. If so, this creates disparate-impact and disparate-treatment liability.
• Collaborate with AI vendors and IT to make sure you have a complete and full understanding of what the AI is doing and how it is evaluating applicants or employees.
• Pay close attention to indemnification provisions in contracts with AI vendors. Vendor contracts should be clear that the tool has been tested for adverse impact, obligations to disclose algorithm changes, indemnification for discriminatory outputs, and audit cooperation.
• Humans are, and should be, integral parts of all employment decisions. Relying too heavily on AI, without critical thinking along the way, can create additional risk.
• Do not skip the interactive process with the employee. AI cannot replace the employer’s legal obligations to engage in the interactive process.
• Train human resources professionals and managers on allowable uses of AI, and maintain a compliant AI policy.

Please reach out to your Stradling attorney if you have any questions, or need assistance with your businesses’ compliance with the ever changing California employment laws.