Disparate impact from AI-driven performance review

Does an AI performance review tool create disparate impact risk?

Yes. The risk starts when AI review outputs affect pay, promotion, discipline, PIP placement, ranking, or termination because the tool can function like an employment selection procedure.

The federal architecture is familiar. It just has new inputs.

Authority	Core rule	Why it matters for performance AI
Title VII, 42 U.S.C. § 2000e-2(k)	A plaintiff identifies a particular employment practice causing disparate impact; the employer then has to show it is job related and consistent with business necessity	Performance-review outputs become relevant when they shape compensation, promotion, discipline, or discharge
UGESP, 29 C.F.R. Part 1607	Employment procedures are measured for adverse impact and validated through criterion-related, content, or construct methods	An AI score or ranking can be treated like a test or other selection procedure even if the interface is modern
Executive Order 14179 and the reported EEOC shift after April 23, 2025	Federal agencies were directed to deprioritize disparate-impact enforcement, but the statute and private right of action were not repealed	The risk profile shifts from agency-led scrutiny toward private litigation and state overlays rather than disappearing

The operative statutory text is still the center of gravity. Title VII asks whether a plaintiff can isolate a particular employment practice causing a protected-class disparity and, if so, whether the employer can show the practice is job related for the position in question and consistent with business necessity. That framework was built in the older testing cases, including Griggs v. Duke Power Co., 401 U.S. 424 (1971), Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975), Watson v. Fort Worth Bank & Trust, 487 U.S. 977 (1988), and Connecticut v. Teal, 457 U.S. 440 (1982). The cases were not about LLMs. That is partly why they still matter. They treat evaluation tools functionally, not by software label.

The Guidelines do the same thing in regulatory language. They apply to any measure, combination of measures, or procedure used as a basis for employment decisions and use “less than four-fifths” as the conventional adverse-impact screen. That matters because the same review system often sits upstream of calibration meetings, bonus pools, PIP entry, layoff rankings, and promotion slates. Once it does, it is difficult to say the tool is only note-taking software. It starts to look more like a scored procedure that affects the terms, conditions, or privileges of employment.compensation, terms, conditions, or privileges of employment

The firms are more aligned than the product market is. Miller Canfield and Hunton Andrews Kurth both read AI evaluation systems through the ordinary Title VII and UGESP frame. Their starting point is not that AI requires a new discrimination doctrine. It is that a review tool becomes an employment procedure once it is used to make decisions about advancement, pay, discipline, or exit. That is a useful simplification. It means the first question is usually classification, not novelty.

The first consequence is classification. Performance-review software stops being just HR software once its outputs are used to rank people, set pay, decide who is promotable, decide who enters a performance process, or help decide who leaves. That shift is functional, not technical. A model that summarizes notes may stay outside the selection-procedure frame. The same model, once its summary is scored or compared in calibration, may not.

Sources for this answer

Primary law

A.1 42 U.S.C. § 2000e-2(k)

Under 42 U.S.C. § 2000e-2(k), an unlawful employment practice is established when a plaintiff demonstrates that a protected characteristic was a motivating factor for an employment decision, regardless of other contributing factors.

an unlawful employment practice is established when the complaining party demonstrates that race, color, religion, sex, or national origin was a motivating factor for any employment practice, even though other factors also motivated the practice.

See 42 U.S.C. § 2000e-2(k).

Primary law

A.6 29 C.F.R. Part 1607

Supports the cited proposition. (29 C.F.R. Part 1607)

less than four-fifths

See 29 C.F.R. Part 1607.

Case law

A.2 Griggs v. Duke Power Co., 401 U.S. 424 (1971)

Under Title VII of the Civil Rights Act of 1964, employment practices that are neutral on their face but have a discriminatory impact are prohibited unless the employer can demonstrate that the practice is a business necessity related to job performance.

The Act proscribes not only overt discrimination but also practices that are fair in form, but discriminatory in operation. The touchstone is business necessity.

See Griggs v. Duke Power Co., 401 U.S. 424 (1971).

Case law

A.3 Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975)

Under Title VII, backpay is an equitable remedy that should be awarded to make victims of discrimination whole unless special circumstances render such an award unjust, and the absence of employer bad faith is not a sufficient reason to deny such relief.

It is true that backpay is not an automatic or mandatory remedy; like all other remedies under the Act, it is one which the courts “may” invoke.

See Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975).

Case law

A.4 Watson v. Fort Worth Bank & Trust, 487 U.S. 977 (1988)

The Supreme Court held that Title VII disparate impact analysis is applicable to subjective or discretionary employment practices, provided the plaintiff identifies the specific practice and demonstrates a causal connection to a significant statistical disparity.

We conclude, accordingly, that subjective or discretionary employment practices may be analyzed under the disparate impact approach in appropriate cases.

See Watson v. Fort Worth Bank & Trust, 487 U.S. 977 (1988).

Case law

A.5 Connecticut v. Teal, 457 U.S. 440 (1982)

Under Title VII, an employer cannot assert a 'bottom-line' defense to justify discriminatory employment practices by pointing to a racially balanced final outcome, as the statute protects individual employees against discriminatory barriers regardless of the overall workforce composition.

We hold that the “bottom line” does not preclude respondent employees from establishing a prima facie case, nor does it provide petitioner employer with a defense to such a case.

See Connecticut v. Teal, 457 U.S. 440 (1982).

Primary law

A.7 EEOC, Questions and Answers to Clarify and Provide a Common Interpretation of...

The Uniform Guidelines on Employee Selection Procedures provide a common interpretation of federal equal employment opportunity law, requiring employers to validate selection procedures only when they result in an adverse impact on protected groups.

The material included is intended to interpret and clarify, but not to modify, the provisions of the Uniform Guidelines.

See EEOC, Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines.

Commentary

A.8 Miller Canfield commentary

Employers may be held liable under Title VII for the use of algorithmic decision-making tools if such tools result in a disparate impact on protected groups and cannot be justified as job-related and consistent with business necessity.

Title VII of the Civil Rights Act of 1964 generally prohibits employers from using neutral tests or selection procedures that have the effect of disproportionately excluding persons based on race, color, religion, sex, or national origin

See Miller Canfield, Adverse Impacts: AI Employment Procedures Under Title VII.

Law-firm commentary

A.9 Hunton Andrews Kurth commentary

The EEOC's recent guidance clarifies that existing federal employment discrimination laws, including the Uniform Guidelines on Employee Selection Procedures, apply to the use of artificial intelligence and algorithmic tools in workplace decision-making.

the EEOC’s Recent Guidance provides broad definitions of key terms—software, algorithms, and artificial intelligence—and explains that the Uniform Guidelines on Employee Selection Procedures (“Guidelines”), which the EEOC adopted way back in 1978

See Hunton Andrews Kurth, EEOC Issues Guidance on Use of AI in Employment Decisions.

Can employers validate an AI performance review tool under federal selection rules?

Sometimes. Employers still need a job-relatedness, validation, subgroup-testing, and documentation story, but dynamic model behavior makes that evidence harder to assemble.

The hard part is not the existence of a validation concept. It is the fit between older validation categories and dynamic models. The Guidelines still organize validation around criterion-related, content, and construct validity. That is straightforward enough for a fixed assessment. It is less clean for an LLM that rewrites review narratives, summarizes free-form feedback, or changes behavior after vendor updates. Perhaps that tension is the central doctrinal fact of the area: the law already knows what kind of question to ask, but the software makes the evidence harder to assemble.

Ogletree Deakins is the clearest on audit methodology. Its 11-step framework treats workplace generative-AI review tools as high-risk when they touch material employment outcomes and pushes the inquiry into a sequence: inventory the use cases, map the data, test for bias, document the results, and review vendor contracts and governance at the same time. That is notable because it looks less like ordinary software procurement and more like a combined employment-law and IO-psych exercise. SIOP's recommendations point in the same direction with more technical language, especially around subgroup testing and differential validity, meaning a tool that predicts performance reasonably well for one group but poorly for another.

We think this is the deepest technical question in the doctrine. UGESP assumes validation studies that are legible and relatively stable. Modern models drift, update, and recombine unstructured inputs. The SIOP material suggests that difficulty does not remove the expectation of validation; it may just make the evidence burden harder to satisfy.

Sources for this answer

Primary law

B.1 29 C.F.R. Part 1607

Supports the cited proposition. (29 C.F.R. Part 1607)

less than four-fifths

See 29 C.F.R. Part 1607.

Primary law

B.2 EEOC, Questions and Answers to Clarify and Provide a Common Interpretation of...

The material included is intended to interpret and clarify, but not to modify, the provisions of the Uniform Guidelines.

See EEOC, Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines.

Commentary

B.3 SIOP, Considerations and Recommendations for the Validation and Use of AI-Based Assessments

AI-based employment assessments must adhere to the same rigorous validation and fairness standards as traditional employment tests while ensuring all development and scoring processes are documented for auditing purposes.

AI-based assessments used to make hiring and promotion decisions require the same level of scrutiny and should meet the same standards that traditional employment tests have been subjected to for decades.

See SIOP, Considerations and Recommendations for the Validation and Use of AI-Based Assessments.

Law-firm commentary

B.4 Ogletree Deakins commentary

Organizations should implement regular, comprehensive AI audits and cross-functional governance frameworks to mitigate legal risks, ensure regulatory compliance, and maintain transparency in their use of AI tools.

As organizations increasingly integrate generative AI tools into daily operations, particularly in HR, AI audits are increasingly important for mitigating legal, operational, and reputational risks.

See Ogletree Deakins, 11 Steps for Performing a Workplace Generative AI Audit.

Does human review reduce legal risk from AI performance scores?

Only if the human review is real. A manager who independently evaluates and rejects model output helps more than a manager who merely confirms AI-drafted ratings or summaries.

The narrower defense remains alive, but only at the margins. Morgan Lewis tries to preserve a meaningful distinction between AI as decision-support and AI as decision-maker. That distinction is plausible where a tool only surfaces information and the manager genuinely evaluates it independently. The same source set also shows why the defense may not travel far. Lattice now markets AI-generated review drafting, real-time performance workflows, and writing assistance directly inside the review chain. Once the machine writes the summary, frames the evidence, and suggests the score, the human decision can start to look more like confirmation than independent judgment.

The second consequence is that human in the loop buys less separation than it first sounds like. A human override can matter. But if the model drafts the review, suggests the rating, highlights the evidence, and anchors the conversation, the legal and practical work may still be getting done by the machine. The manager's presence then changes the workflow more than it changes the causal story. That is why the more serious commentary keeps circling back to validation, subgroup testing, and audit trails rather than treating supervision alone as a cure.

The third consequence is that product design and product positioning now matter more than they used to. Microsoft's Viva documentation says the product is “not designed to enable employee evaluation, tracking, automated decision making, profiling, or monitoring”. Lattice, by contrast, markets AI directly into appraisal workflows. That does not decide the case by itself, but it changes how easy it is to argue that the tool was merely adjacent to evaluation rather than built to influence it.

Perhaps some review layers will matter, especially where managers genuinely re-score or reject model output. But the source set also points to automation bias and product designs that make the machine draft the starting point of the human judgment. The line between assistance and substantial influence is still fact-bound rather than settled.

Sources for this answer

Law-firm commentary

C.1 Morgan Lewis commentary

Employers must maintain human oversight and responsibility for workplace safety when utilizing AI tools, as these technologies function as decision-support systems rather than legal substitutes for an employer's non-delegable duty to protect workers.

AI systems increasingly used in safety programs—such as platforms that flag high-risk activities, suggest mitigation measures, or auto-generate safety documentation—are best understood as decision-support tools, not decision-makers.

See Morgan Lewis, Using AI to Improve Safety: Managing the Legal Risks Alongside the Benefits.

Commentary

C.2 Lattice, February 2026 Product Updates

Lattice's February 2026 product updates introduce enhanced administrative controls for compensation cycles, performance review visibility, and granular data permissions to improve operational flexibility and reduce manual workflows.

Recalculate Budget & Guidance During Active Cycles: Receive proactive alerts when eligibility changes impact budgets, and recalculate affected guidance in one click.

See Lattice, February 2026 Product Updates.

Commentary

C.3 Lattice, Using AI to Write Performance Reviews: Everything You Need to Know

While generative AI can assist in drafting performance reviews by providing structure and summarizing data, organizations must implement ethical guardrails, ensure human oversight, and protect employee privacy to mitigate risks such as bias, data breaches, and loss of trust.

The key to the ethical use of AI within the review process is “to think of AI as your assistant, not your replacement,” recommended Fesinstine.

See Lattice, Using AI to Write Performance Reviews: Everything You Need to Know.

Commentary

C.4 Lattice, How to Evaluate and Choose Employee Appraisal Software

Modern employee appraisal software utilizes AI-driven features and structured data management to improve the objectivity, consistency, and efficiency of performance reviews compared to traditional manual methods.

Employee appraisal software is designed to track and improve the appraisal process. The right system helps organisations move from static, ineffective appraisals toward a dynamic, effective AI-enabled process.

See Lattice, How to Evaluate and Choose Employee Appraisal Software.

Law-firm commentary

C.5 Ogletree Deakins commentary

As organizations increasingly integrate generative AI tools into daily operations, particularly in HR, AI audits are increasingly important for mitigating legal, operational, and reputational risks.

See Ogletree Deakins, 11 Steps for Performing a Workplace Generative AI Audit.

Commentary

C.6 SIOP, Considerations and Recommendations for the Validation and Use of AI-Based Assessments

AI-based assessments used to make hiring and promotion decisions require the same level of scrutiny and should meet the same standards that traditional employment tests have been subjected to for decades.

See SIOP, Considerations and Recommendations for the Validation and Use of AI-Based Assessments.

Vendor documentation

C.7 Microsoft Learn, Introduction to Viva Insights

Supports the cited proposition. (Microsoft Learn, Introduction to Viva Insights)

not designed to enable employee evaluation, tracking, automated decision making, profiling, or monitoring

See Microsoft Learn, Introduction to Viva Insights.

Vendor documentation

C.8 Microsoft, Viva Insights Overview

Microsoft Viva Insights provides organizations with data-driven tools to analyze workplace patterns, employee sentiment, and productivity metrics to improve organizational performance.

Bring together workplace data and employee sentiment to analyze organizational performance.

See Microsoft, Viva Insights Overview.

Can AI performance data create proxy discrimination or pay-equity risk?

Yes. Review data can carry historical promotion, compensation, communications, location, leave, sentiment, and monitoring patterns into decisions even when protected traits are not explicit fields.

Fisher Phillips puts the emphasis on the data story. Its performance-management commentary warns that predictive systems for internal mobility and employee development can simply learn the employer's old promotion and compensation patterns. That is a narrower but probably more practical claim than generic talk about AI bias. If historical promotion notes, prior review language, and pay outcomes were skewed, the model can preserve the old hierarchy while presenting the result as neutral scoring. Hunton makes the adjacent compensation point: once automated decision tools influence wage-setting, review logic and pay-equity logic start to merge.

The fourth consequence is about data rather than scores. Performance systems increasingly pull from communications metadata, goal-tracking tools, collaboration history, prior review language, and sometimes sentiment or monitoring layers. That creates a proxy problem. Protected traits may be absent from the schema and still reappear through leave patterns, geography, language style, meeting load, school history, or historical manager comments. Box's governance materials make the adjacent point in operational language: broad AI access rights tend to pull in more information than the user remembers is there. The measurement problem can therefore arrive before anyone sees a protected-class field.

The fifth consequence is that compensation is not a separate system for long. Performance scores frequently feed bonus allocations, merit increases, or promotion timing. When that happens, a biased review model becomes a pay model without a formal handoff. That is why the source set's wage-setting commentary sits naturally beside the review commentary. If historical compensation or promotion outcomes are part of the training or recommendation logic, the system can inherit old disparities at the exact point where companies think they are merely rewarding current performance. The Equal Pay Act questions in the research set push in the same direction, especially where salary-history logic is still embedded somewhere in the stack.

The sixth consequence is that not every AI-related performance tool belongs in the same bucket. That cuts against panic as much as it cuts against easy deployment. A wellbeing or productivity analytics tool that is structurally kept out of individual evaluation may stay closer to Microsoft Viva's posture. A system that generates review language, internal mobility rankings, or score recommendations looks different. Public company statements are starting to show the pressure from the other side as well. Shopify's AI-first posture treats AI use as a baseline work expectation. Once companies begin measuring AI proficiency in the review system itself, the doctrine will likely move faster, because the connection between tool use and employment consequence becomes explicit rather than inferred.

Perhaps in some narrow roles, some of these signals will correlate with legitimate performance questions. The source set still reads skeptical overall. Research on sentiment tools and algorithmic monitoring suggests measurement bias and organizational drag, not just cleaner evidence about employee performance.

Sources for this answer

Law-firm commentary

D.1 Fisher Phillips commentary

Employers deploying AI for performance management and promotion decisions face significant legal risks related to anti-discrimination compliance, necessitating rigorous data validation, human oversight, and ongoing monitoring to mitigate potential bias.

AI tools used for performance management, promotion recommendations, and skill inference present risk under employment and anti-discrimination laws.

See Fisher Phillips, How Employers Can Manage Risk When Using AI for Employee Performance Management.

Law-firm commentary

D.2 Hunton Andrews Kurth commentary

State legislators are increasingly regulating the use of AI in employee compensation decisions, creating new compliance obligations and legal risks for employers under both existing employment statutes and emerging state-specific AI laws.

state lawmakers are increasingly scrutinizing employers’ use of AI and automated decision tools to set or influence employee compensation

See Hunton Andrews Kurth, State Lawmakers Seek to Regulate Employer Use of AI for Wage Decisions.

Commentary

D.3 Iowa Law Review, Proxy Discrimination in the Age of Artificial Intelligence and Big DataPDF

The authors argue that because AI systems are inherently designed to identify correlations in data, they will inevitably engage in proxy discrimination by using facially neutral data to replicate the predictive power of legally protected characteristics, thereby undermining the effectiveness of existing anti-discrimination regimes.

Proxy discrimination is a particularly pernicious subset of disparate impact. Like all forms of disparate impact, it involves a facially neutral practice that disproportionately harms members of a protected class.

See Iowa Law Review, Proxy Discrimination in the Age of Artificial Intelligence and Big Data.

Commentary

D.4 Harvard Law Review, Resetting Antidiscrimination Law in the Age of AI

Legislative trends in AI regulation, particularly the widespread adoption of output-based auditing and testing requirements, suggest a normative preference for systemic disparate impact remedies that should inform the interpretation of existing antidiscrimination law.

Of the four regulatory methods this Chapter surveys, the most common for addressing AI-enabled bias regulates an AI system’s outputs.15 This method involves requirements to test, audit, report, or adjust AI systems based on their results.

See Harvard Law Review, Resetting Antidiscrimination Law in the Age of AI.

Commentary

D.5 Michigan Technology Law Review, Discrimination by Proxy: How AI Uses Big Data to Discriminate

Existing legal frameworks are insufficient to address the unintentional and sophisticated nature of proxy discrimination in AI systems, necessitating new regulatory and statutory approaches.

AI systems might slip discrimination past current laws through “proxy discrimination” without new regulatory and statutory approaches.

See Michigan Technology Law Review, Discrimination by Proxy: How AI Uses Big Data to Discriminate.

Commentary

D.6 Box AI Principles: Responsible AI Development and Governance

Box maintains a structured AI governance program and set of principles designed to ensure data security, regulatory compliance, and customer control over AI model training and content access.

Box won’t train AI models using customer content without the customer’s explicit authorization

See Box AI Principles: Responsible AI Development and Governance.

Commentary

D.7 Box, AI First Part 1

Effective enterprise AI governance requires a hybrid model where functional leaders own specific transformation initiatives within a centrally defined strategic framework and existing organizational guardrails.

We’ve built a governance model where functional leaders own AI transformation for their teams, operating within a centrally-defined strategy and guardrails, and supported by specialized technical teams who can convert ideas into agentic workflows.

See Box, AI First Part 1.

Law-firm commentary

D.8 Seyfarth Shaw, 2026 Developments In Equal Pay Litigation Book

Federal courts are increasingly adopting a middle-ground approach to Equal Pay Act claims, which permits the use of prior salary as a factor other than sex only when combined with other legitimate factors and provided the prior salary itself was not the product of sex discrimination.

the EPA only precludes an employer from relying solely upon a prior salary to justify pay disparity.

See Seyfarth Shaw, 2026 Developments In Equal Pay Litigation Book.

Law-firm commentary

D.9 Winston & Strawn, Ninth Circuit Rules That Prior Salary Cannot Justify Wage Differential

The Ninth Circuit held that an employer cannot justify a wage differential between male and female employees under the Equal Pay Act by relying on an employee's prior salary history.

the court held that prior salary—alone or in combination with other factors—cannot justify a wage differential between male and female employees.

See Winston & Strawn, Ninth Circuit Rules That Prior Salary Cannot Justify Wage Differential.

Commentary

D.10 GAI Insights, The CEO of Shopify Has a GREAT Memo to Employees about AI

Corporate leadership can effectively integrate artificial intelligence into organizational operations by mandating AI usage, tying AI proficiency to performance evaluations, and requiring AI-first solutions before authorizing additional headcount.

Shopify’s memo makes AI usage mandatory, not optional “extra credit,” for all employees, including leaders.

See GAI Insights, The CEO of Shopify Has a GREAT Memo to Employees about AI.

Primary law

D.11 PMC, Queerphobia in Sentiment Analysis

Automated sentiment analysis tools frequently exhibit predictive bias against marginalized groups, including queer identities, which can undermine the validity of research and lead to representational or allocational harms.

Unfortunately, like many natural language processing techniques, sentiment analysis can show bias against marginalised groups.

See PMC, Queerphobia in Sentiment Analysis.

Commentary

D.12 Cornell Chronicle, More Complaints, Worse Performance When AI Monitors Work

Research indicates that algorithmic surveillance in the workplace can lead to decreased employee performance and increased resistance unless the technology is implemented for developmental purposes that allow for human contextualization.

Organizations using AI to monitor employees’ behavior and productivity can expect them to complain more, be less productive and want to quit more – unless the technology can be framed as supporting their development

See Cornell Chronicle, More Complaints, Worse Performance When AI Monitors Work.

Did the EEOC pullback end AI performance review disparate-impact risk?

No. The April 23, 2025 federal pullback changed enforcement posture, but it did not repeal Title VII, remove private lawsuits, or eliminate state and vendor-facing risk.

April 23, 2025 changed the federal backdrop but perhaps not the rule itself. Executive Order 14179 directed agencies to deprioritize disparate-impact enforcement. The law-firm commentary in the source set also reports a practical EEOC shift toward closing certain disparate-impact matters and issuing right-to-sue notices. But the same commentary is consistent on the limiting point: the order did not rewrite Title VII, repeal the Guidelines, or erase private suits built on the same statutory text. The enforcement mix changed more than the underlying doctrine did.

On enforcement, the firms are converging around a two-track story. K&L Gates, Hinshaw, and Epstein Becker Green all treat the White House and EEOC shift as real, but incomplete. At the same time, Whiteford Taylor Preston, CDF, and Epstein Becker Green point to Mobley v. Workday and Kistler v. Eightfold AI as signs that plaintiffs are also attacking vendors and data workflows, not just employers and end results. Those cases are hiring cases, not performance-review cases. But they still weaken the comforting idea that the vendor owns the hard parts of the model and therefore owns the legal problem too.

Mobley and Kistler are not performance-review decisions. They are still signals. They suggest courts and plaintiffs may look through the old we only sell software framing when the vendor is deeply involved in scoring logic, data assembly, or applicant profiling. It is not yet clear how far that reasoning carries into internal review systems, but it probably weakens the idea that procurement fully transfers the hard parts of the problem.

The source set is fairly consistent that it changed enforcement posture more than statutory substance. The remaining uncertainty is practical rather than doctrinal: whether private plaintiffs, state regulators, and contract disputes end up doing most of the work that agencies might once have done themselves.

Sources for this answer

Primary law

E.1 White House, Restoring Equality of Opportunity and Meritocracy

This executive order establishes a federal policy to eliminate and deprioritize the enforcement of disparate-impact liability theories in civil rights and employment law, asserting that such theories are inconsistent with the Constitution.

It is the policy of the United States to eliminate the use of disparate-impact liability in all contexts to the maximum degree possible to avoid violating the Constitution, Federal civil rights laws, and basic American ideals.

See White House, Restoring Equality of Opportunity and Meritocracy.

Law-firm commentary

E.2 Hinshaw & Culbertson, Presidential Executive Order Seeks to Eliminate Disparate Impact Liability

Although a presidential executive order may deprioritize federal enforcement of disparate impact theories, it does not amend Title VII or invalidate the underlying judicial precedent that permits private disparate impact claims.

While the EO signals a major policy change, it does not alter the underlying statutes or U.S. Supreme Court precedent that recognize disparate impact as a valid legal theory.

See Hinshaw & Culbertson, Presidential Executive Order Seeks to Eliminate Disparate Impact Liability.

Law-firm commentary

E.3 Epstein Becker Green commentary

Despite potential shifts in EEOC enforcement priorities, employers remain subject to disparate impact liability for unintentional discrimination through private litigation and state or local regulatory compliance.

Employers may be liable when they use AI that algorithmically discriminates, even if done so unintentionally.

See Epstein Becker Green, Artificial Intelligence and Disparate Impact Liability: How the EEOC's End to Disparate Impact Claims Affects Workplace AI.

Law-firm commentary

E.4 K&L Gates commentary

The January 2025 executive order on artificial intelligence revoked the Biden administration's EO 14110, leading federal agencies to rescind or update AI-related guidance while shifting the regulatory focus toward state-level compliance.

President Trump issued the AI EO to (1) implement the revocation of President Biden’s executive order on artificial intelligence (AI), entitled the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (EO 14110)

See K&L Gates, The Changing Landscape of AI: Federal Guidance for Employers Reverses Course with New Administration.

Law-firm commentary

E.5 Whiteford Taylor Preston commentary

Employers face increasing legal liability for AI hiring tools as courts classify vendors as agents or consumer reporting agencies, necessitating robust vendor due diligence and internal governance infrastructure.

The plaintiffs’ theory rests on the Fair Credit Reporting Act (FCRA), which mandates specific procedures, including disclosure, access, and the opportunity to dispute errors, when companies compile “consumer reports” for employment decisions.

See Whiteford Taylor Preston, Employment Law Update: AI Hiring Under Fire: Algorithmic Screening Enters The Chat.

Commentary

E.6 CDF Labor Law, AI Lawsuit Pushes the Boundaries of AI Litigation and May Signal a New Wave

Existing statutes like the Fair Credit Reporting Act may be applied to modern AI hiring tools, creating potential liability for employers even when they do not directly control the underlying algorithms.

The Eightfold AI case underscores a critical point: even decades-old statutes not written with AI in mind can be applied to modern technology.

See CDF Labor Law, AI Lawsuit Pushes the Boundaries of AI Litigation and May Signal a New Wave.

Law-firm commentary

E.7 Epstein Becker Green, AI Hiring Tools and Consumer Reports: Understanding the Eightfold Litigation

The Eightfold litigation highlights potential legal risks for employers using AI-driven hiring tools, specifically regarding whether such tools and their generated scores constitute consumer reports subject to the FCRA and ICRAA.

the lawsuit, which is pending in California’s Superior Court for the County of Contra Costa, alleges that Eightfold violated the FCRA by failing to meet the certification, notification, disclosure, authorization, and dispute requirements

See Epstein Becker Green, AI Hiring Tools and Consumer Reports: Understanding the Eightfold Litigation.