The Algorithm Is Regulated. The Executive Is Not.

The EEOC regulates the algorithm. No agency regulates the executive who fires based on it.

In March 2023, the Equal Employment Opportunity Commission published technical assistance on artificial intelligence and algorithmic fairness under Title VII of the Civil Rights Act, the Americans with Disabilities Act, and the Age Discrimination in Employment Act. The guidance, issued through the Commission’s Artificial Intelligence and Algorithmic Fairness Initiative, addressed AI tools used in employment selection — résumé screening, skills assessments, video-interview scoring. Under the framework, an employer using an AI-powered hiring tool bears the same adverse-impact liability it would bear if a human interviewer made the discriminatory selection. The four-fifths rule applies: if the selection rate for a protected group falls below 80 percent of the rate for the most-selected group, the employer must validate the tool or face Title VII exposure.

The instrument is strong at its designed scale. It extends a well-established disparate-impact framework — Section 703(k) of Title VII, operationalized through the Uniform Guidelines on Employee Selection Procedures at 29 C.F.R. Part 1607 — to algorithmic decision-makers without requiring new legislation. The Commission identified concrete failure modes: AI trained on historically discriminatory hiring data that replicates past bias; automated systems screening out applicants with disabilities who do not perform well on standardized assessments; age-biased algorithms downgrading older workers. The employer is liable for what the tool does. The Commission did not wait for Congress. It read the statute it had and applied it.

That is the instrument at its best — a regulatory body extending a proven anti-discrimination framework to a new technology using existing statutory authority. The problem is that it addresses one slice of the AI-employment problem and does not reach the larger pattern the current hiring data has exposed.

U.S. public companies shrank white-collar workforces through 2024 and into 2025 on the assumption that AI could shoulder more tasks. That assumption was not an algorithm applied to a résumé pile. It was a boardroom judgment about how many humans the company would need. The EEOC’s guidance governs the algorithm’s interaction with a job applicant. It does not govern the executive’s decision to eliminate a position because software might eventually perform it. The Worker Adjustment and Retraining Notification Act — WARN, 29 U.S.C. §§ 2101–2109 — requires 60 days’ notice for mass layoffs of 100 or more employees, but governs the notification event, not the workforce-structuring judgment that precedes it. An employer that eliminates 99 positions escapes WARN. An employer that eliminates 500 positions over six months through rolling programs may also escape, depending on how the cuts are timed and structured. The statute does not ask whether the layoff was based on an AI projection that later proved wrong.

The companies reversing course this week make the gap concrete. Booz Allen Hamilton cut thousands of jobs as federal contracts shrank, then told investors it is “a little bit behind” on hiring and needs to accelerate. Chief Operating Officer Kristine Martin Anderson’s acknowledgment that the company over-cut is an admission that the workforce-elimination judgment was wrong. The workers who were cut are not necessarily the workers being rehired. The restructuring has already occurred. The displaced workforce has already absorbed the cost. No federal instrument governed the decision to eliminate those positions based on AI projections, and no federal instrument governs the aftermath.

Alphabet expects to continue hiring in AI and cloud. CSX plans to increase train and engine head count. ServiceNow wants more sales executives. The pattern is consistent across industries: companies that reduced head count on AI-capability projections are rehiring because the projections were wrong. The rehiring does not undo the displacement. The workers who were laid off or never hired have already moved, retrained, or absorbed the gap in employment that no governing instrument was designed to prevent.

Sarah Franklin, CEO of the HR platform Lattice, told the Journal that companies stopped hiring entry-level employees, thinking AI agents could replace them. Her clients are back in hiring mode for junior positions — the roles that were supposed to be the first to go. “Just because you have coding agents doesn’t mean you’re not hiring engineers,” she said. The elimination of those positions was not an adverse-selection event the EEOC’s guidance was designed to catch. It was a strategic judgment about the future of work, imposed on workers who had no voice in the judgment and no legal remedy when it reversed.

Congress has not acted. The Algorithmic Accountability Act, first introduced in 2019 and reintroduced in subsequent sessions, would require impact assessments for automated employment decision systems. It has not passed. The European Union’s Artificial Intelligence Act, finalized in 2024, classifies AI systems used in employment — including workplace management and termination decisions — as high-risk and imposes transparency, human-oversight, and conformity-assessment requirements. The United States has no equivalent. The workforce development infrastructure funded by the Workforce Innovation and Opportunity Act, reauthorized in 2014, was built for displacement events that last years and move in one direction. It was not built for executive judgments about AI capability that reverse in months.

Urban labor markets concentrate the cost. The workers displaced by AI-projection-based restructuring are disproportionately in metropolitan areas where white-collar employment is dense. When Booz Allen cuts from its Virginia workforce or Alphabet freezes entry-level hiring at its Mountain View campus, the displacement lands in the same metropolitan labor markets served by the federal workforce development system — state workforce agencies, local workforce development boards, American Job Centers. That infrastructure absorbs displaced workers into retraining and placement services, but it was designed for a transition measured in years. The current pattern is faster, driven by executive projections that the executives have now abandoned, and concentrated in the cities that house the headquarters where the projections were made.

Robert Half CEO M. Keith Waddell told investors that AI’s impacts on the job market are proving “more benign than some have feared.” Jobless claims are at their lowest since 1969. The aggregate is benign. The specific is not: workers laid off or never hired based on AI projections the company has now admitted were wrong, with no governing instrument that reached the decision when it was made and no remedy that reaches it now.

The EEOC built a strong instrument for the problem it was designed to solve. The problem it was not designed to solve — executives eliminating jobs based on AI capability projections that prove wrong — is larger, more consequential, and entirely ungoverned. Congress has had six years. The EU acted. The United States has not.