Category: AI Issues

By Gerald L Maatman, Jr., Shannon Noelle, and Elizabeth G. Underwood

Duane Morris Takeaways: On November 20, 2025, in Buchanan v. Vuori, Inc., No. 5:23-CV-01121 (N.D. Cal. Nov. 20, 2025), Magistrate Judge Nathanael M. Cousins of the U.S. District Court for the Northern District of California imposed sanctions on plaintiff’s counsel for using artificial intelligence to generate case law citations in a motion for preliminary approval of a wage and hour collective action settlement.  The sanctions included an order directing plaintiff’s counsel to pay $250 to the clerk of court, striking the motion without leave to refile, and referring plaintiff’s counsel to the Court’s Standing Committee on Professional Conduct.  Importantly, because of the sanctions, Magistrate Judge Cousins found plaintiff’s counsel to be an inadequate representative of the class and precluded plaintiff’s counsel from filing an additional motion for approval of the class settlement.  This required defense counsel to file a case management statement requesting a stipulation of dismissal that was approved on January 8, 2026.  Plaintiff’s counsel’s use of AI ultimately delayed final disposition of the action until months later and underscores the growing trend of judicial commitment to accountability with respect to attorney use of AI in drafting legal filings.

Case Background

On March 14, 2023, a former Vuori, Inc. (“Vuori”) employee, Terrence Buchanan, sued Vuori, alleging that it had violated the Fair Labor Standards Act (FLSA) and various California Labor Codes by miscalculating the overtime paid to their employees by failing to include commissions or bonuses in calculating overtime.  See Case No. 5:23-cv-01121, ECF No. 1. Eventually, the parties settled the litigation.

On October 3, 2025, after a first try for settlement approval failed, counsel for Plaintiff filed a second motion for preliminary approval of a collective action settlement (ECF No. 81) followed by a corrected motion on October 28, 2025 (ECF No. 89).  Upon review of the corrected motion, the Court found that the memorandum in support of the motion included 8 quotations “supposedly attributable to a real case” that did not actually appear in the cited case and “one nonexistent case.”  See ECF No. 96, at 1.  On November 5, 2025, the Court ordered plaintiff’s counsel to show cause as to why he should not be sanctioned pursuant to Federal Rule of Civil Procedure 11(c) and referred to the Court’s Standing Committee on Professional Conduct under Civil Local Rule 11-6 for providing fabricated case law to the Court.  Plaintiff’s counsel filed a response and proof of service that he provided the Court’s order to show cause to his client.  See ECF Nos. 92, 93.  He also filed a supplemental response.  See ECF No. 94.  The Court held a hearing on the order to show cause on November 19, 2025, at which counsel and plaintiff Buchanan appeared.  See ECF No. 96, 1-2.

Order Imposing Sanctions And Finding Class Counsel Is Therefore Inadequate  

Magistrate Judge Cousins ordered sanctions by way of payment of $250 to the clerk of court pursuant to Federal Rule of Civil Procedure 11(c), referred Plaintiff’s counsel to the Standing Committee on Professional Conduct pursuant to Civil Local Rule 11-6, and ordered that the motions for preliminary approval be stricken without leave to refile. 

In support of this decision, Magistrate Judge Cousins explained that “the rise in non-existent cases and quotations hallucinated by artificial intelligence tools” is of “particular concern.”  ECF No. 96 at 3.  He noted that Plaintiff’s counsel “acknowledge[d] without reservation” that his motion “contained one non-existent case citation.”  ECF No. 92, at 3 (citing ECF No. 92 at 2).  Plaintiff’s counsel also admitted to using about six different AI tools to prepare his motion “[a]s a solo practitioner under time pressure” and that he used the tools to check one another.  Id. at 3-4.  The Court noted that the corrected memorandum of law in support of the second motion for preliminary approval, did not correct the false case law hallucinated by the AI tools.  Id. at 4.  The Court made clear that the intentions of Plaintiff’s counsel were irrelevant and that his use of AI which “led him to submit a hallucinated case to the Court through his motion” and failure to conduct a reasonable inquiry into the law cited in his motion violated Rule 11(b) and Local Rule 11-4.  Id. at 4-5.  Specifically, the Court found that Plaintiff’s counsel violated his duty of candor owed to the tribunal under California Rule of Professional Conduct 3.3 by citing nonexistent cases and quotations to the Court and certifying “via signature that he had conducted reasonable inquiry into these citations when he had not.”  Id. at 5.

Though Plaintiff’s counsel offered to forfeit attorneys’ fees in the matter, to file an amended motion certifying that he verified all citations, and to complete continuing legal education, the Court declined his suggested sanctions and instead ordered that:  (1) plaintiffs’ second motion for preliminary approval of a class action settlement and corrected motion be stricken with prejudice; (2) Plaintiff’s counsel pay the clerk of court $250 by December 5, 2025; and (3) Plaintiff’s counsel be referred to the Court’s Standing Committee on Professional Conduct in connection with his violation of Local Rule 11-4 and unprofessional conduct.  As to the third remedial measure, the Standing Committee has authority to conduct further investigation or impose additional discipline, such as continuing legal education or notification of the state bar as it deems necessary and appropriate.  Magistrate Judge Cousins added that it was the Court’s “hope” that “the experience with the Standing Committee also proves constructive for Plaintiff’s counsel, who attests that he is a very busy sole practitioner who faces various logistical constraints.”  See ECF No. 96 at 6. 

Finally, and notably, the Court found that striking plaintiff’s motion for settlement approval “necessarily raises the questions” of whether Plaintiff’s counsel could adequately represent the class through final approval of settlement.  The Court found that Plaintiff’s counsel could not file an amended motion for preliminary approval of the class settlement because “it does not find that he is adequate class counsel, which would prevent the Court from approving a renewed motion for settlement approval.”  See ECF No. 96, at 7.

Delay Of Final Disposition Due To Sanctions And Inadequate Class Representative Finding  

On December 5, 2025, the Court docketed and acknowledged receipt of counsel’s payment of $250 to the clerk of court.  See ECF No. 98.  As Magistrate Judge Cousins found Plaintiff’s counsel to be an inadequate class representative and therefore prohibited him from filing further motions to approve the class action settlement, on January 7, 2026, counsel for Vuori was required to file a case management statement to get final disposition of the action and setting out Vuori’s position that the parties signed a settlement agreement containing “a general release of Plaintiff’s claims against Defendant” and, per the terms of that agreement, “Plaintiff was obligated to dismiss this action with prejudice no later than December 31, 2025.”  See ECF No. 100, at 2.  To that end, counsel for Vuori requested that “Plaintiff immediately dismiss this action with prejudice.”  Id.  On that same day, Plaintiff’s counsel filed a Stipulation of Dismissal with the Court.  See ECF No. 101.  On January 8, 2026, the Court granted the stipulation of dismissal with prejudice by order signed by Magistrate Judge Cousins.  See ECF No. 102.

Implications For Companies

This order is unprecedented. The implications of the sanctions order and the aftermath of the order is two-fold.  First, employers and companies should review class counsel’s filings scrupulously by noting any citations or quotations that seem incorrect and AI-generated as this may build a case for disqualifying class counsel and may prove as a barrier to getting approval of a class settlement agreement.  Second, employers and companies must be diligent in ensuring that in-house and outside counsel alike use human verification in connection with the use of any AI tool when drafting court filings to ensure that all case law citations and quotations have been independently verified by an attorney prior to filing such information with a court to avoid similar deleterious consequences.   

By Gerald L. Maatman, Jr., Justin R. Donoho, and Hayley Ryan

Duane Morris Takeaways:  On December 11, 2025, President Donald J. Trump signed Executive Order 14365 titled “Ensuring a National Policy Framework for Artificial Intelligence.” The Order targets what it characterizes as a “patchwork” of State-by-State AI regulation and directs federal agencies to pursue a more uniform, national framework. Rather than serving as a technical AI governance roadmap, the Order focuses on limiting State AI laws through federal funding leverage, potential preemption, and expanded use of FTC enforcement authority. The discussion below highlights the Order’s core objectives and key implications for companies and employers. The Executive Order is required reading for any organizations deploying AI or thinking of doing so.

The Executive Order’s Core Objectives

Reduce State AI Regulation By Framing It As A Competitiveness Problem

The Order emphasizes U.S. leadership in artificial intelligence and asserts that divergent State regulatory regimes increase compliance costs, especially for startups, and may impede innovation and deployment. It also raises concerns that certain State approaches could pressure companies to embed “ideological” requirements into AI systems.

Create Leverage Through Federal Funding: BEAD Broadband Money As The “Carrot And Stick”

Within 90 days, the Secretary of Commerce is directed to issue a policy notice describing the circumstances under which States may be deemed ineligible for certain broadband deployment funding under the Broadband Equity Access and Deployment (BEAD) program if they impose specified AI-related requirements. The notice is also intended to explain how fragmented State AI laws could undermine broadband deployment and high-speed connectivity goals.

Move Toward A Federal Reporting And Disclosure Standard

Within 90 days after the Order’s State-law “identification” process (discussed below), the Federal Communications Commission (FCC), in consultation with a Special Advisor for AI and Crypto, is instructed to consider whether to initiate a proceeding to adopt a federal reporting and disclosure standard for AI models that would preempt conflicting State requirements.

Use The FTC Act As An Enforcement Anchor And Tee Up Preemption Arguments

Within 90 days, the Federal Trade Commission (FTC) is directed, in consultation with other federal agencies, to issue a policy statement addressing how the FTC Act’s prohibition on unfair or deceptive acts or practices applies to AI models, with the express objective of preempting conflicting State laws.

Establish A Federal AI Litigation Task Force To Challenge State AI Laws

The Executive Order goes beyond policy statements and funding leverage by directing the Attorney General, within 30 days, to establish an AI Litigation Task Force dedicated exclusively to challenging State AI laws that conflict with the Order’s national policy objectives. The Task Force is authorized to pursue constitutional and preemption-based challenges, signaling an intent to bring coordinated, affirmative litigation against State AI regimes.

That enforcement effort is reinforced by a parallel State-law triage process. Within 90 days, the Secretary of Commerce must publish an evaluation identifying “onerous” State AI laws for potential challenge, particularly those that require AI systems to alter truthful outputs or compel disclosures that may implicate First Amendment or other constitutional concerns. Together, these provisions signal an intent to move quickly from policy articulation to test cases aimed at curbing State-level AI regulation.

Implications For Companies

Compliance Strategy May Shift, But Uncertainty Rises First

Although companies may welcome relief from conflicting State AI mandates, the Executive Order is likely to increase near-term uncertainty. Preemption disputes are likely, and the Order directs agency action rather than establishing a comprehensive statutory framework. Companies should avoid scaling back State-law compliance prematurely and should assume any federal override will be contested until resolved through rulemaking and litigation.

Class Action Exposure Will Shift, Not Disappear

Even if State AI laws are narrowed, plaintiffs’ lawyers are likely to pursue claims under more traditional theories, including consumer protection (particularly AI marketing and disclosure claims), employment discrimination, privacy and biometrics statutes, and contract or misrepresentation theories. The Order’s emphasis on FTC unfair and deceptive practices enforcement suggests that federal consumer protection standards may become the new focal point for both regulatory scrutiny and follow-on civil litigation.

Employment Risk Remains

Employers should expect ongoing scrutiny of AI use in hiring, promotion, and performance management, including disparate impact claims, vendor-liability arguments, and discovery disputes over model documentation, adverse impact analyses, and validation. Defensible governance, testing, and documentation remain critical.

Federal Contracting And Funding May Come With New AI Representations

If federal agencies adopt standardized AI disclosures, companies operating in regulated industries or participating in broadband initiatives may face new contract provisions governing AI use, along with enhanced reporting and audit obligations.

What Companies Should Do Now

Companies should begin by identifying where and how AI tools are being deployed, particularly in consumer-facing and employment-related contexts, and evaluating those uses under existing disclosure, privacy, and anti-discrimination laws. Public-facing statements about AI capabilities should be reviewed to ensure they are accurate and defensible, as increased regulatory and litigation focus on unfair or deceptive practices is likely to heighten scrutiny of AI-related claims. Companies should also review vendor relationships to confirm that contracts clearly address testing and validation obligations, incident response, audit rights, and appropriate allocation of risk for privacy and discrimination claims. Finally, organizations should remain prepared for continued regulatory change by maintaining State-law compliance readiness while monitoring federal agency actions that may shape a national AI framework.

Bottom Line

This Executive Order is a significant policy signal. The federal government is positioning itself to reduce State-by-State AI regulation and replace it with a framework centered on federal disclosure requirements and consumer protection enforcement. Companies should view the Order as an opportunity to prepare for a likely federal compliance baseline, without assuming State-law exposure will disappear in the near term.

By Justin Donoho

Duane Morris Takeaway: Available now is the recent article in the Legal Intelligencer by Justin Donoho entitled “Gen AI Class Action Key Decisions and Trends in 2025.”  The article is available here and is a must-read for corporate counsel involved with gen AI technologies.

This year has been a busy one in the generative artificial intelligence (gen AI) class action litigation landscape. New pleadings were filed, including several new class actions, several consolidated and amended complaints, and one appeal.  Several key decisions were issued, including a trio that formed a three-way split of authority on how to determine whether training a gen AI model on copyrighted materials constitutes “fair use” under the Copyright Act.  Additionally, one humongous settlement was reached.  Additional notable decisions issued in 2025 in gen AI class actions include a decision denying class certification on the basis of the class definition being defined as a “fail-safe” class, dispositive decisions defining the contours of claims alleging that gen AI developers violated the Digital Millenium Copyright Act, a decision on the copyrightability of voice in the context of voice cloning technology, and multiple additional decisions on motions to compel, further clarifying the scope of documents that may or may not be discoverable in gen AI class actions.  This article analyzes these key decisions and trends.

Implications For Corporations

With gen AI continuing to proliferate and the current presidential administration continuing the prior administration’s policy goals of sustaining and enhancing America’s global AI dominance, gen AI litigation is multiplying. We should expect to see an upward trend of key decisions and new cases in the remainder of this year and beyond as this burgeoning area of the law continues to unfold.

By Gerald L. Maatman, Jr., Justin Donoho, and Hayley Ryan

Duane Morris Takeaways: In a series of discovery rulings in the case of In Re OpenAI, Inc. Copyright Infringement Litigation, No. 23 Civ. 11195 (S.D.N.Y.), Magistrate Judge Ona T. Wang issued a series of orders that signal how courts are likely to approach AI data, privacy, and discovery obligations. Judge Wang’s orders illustrate the growing tension between AI system transparency and data privacy compliance – and how courts are trying to balance them.

For companies that develop or use AI, these rulings highlight both the risk of expansive preservation demands and the opportunity to share proportional, privacy-conscious discovery frameworks. Below is an overview of these decisions and the takeaways for in-house counsel, privacy officers, and litigation teams.

Background

In May 2025, the U.S. District Court for the Southern District of New York issued a preservation order in a copyright action challenging the use of The New York Times’ content to train large language models. The order required OpenAI to preserve and segregate certain output log data that would otherwise be deleted. Days later, the Court denied OpenAI’s motion to reconsider or narrow that directive. By October 2025, however, the Court approved a negotiated modification that terminated OpenAI’s ongoing preservation obligations while requiring continued retention of the already-segregated data.

The Court’s Core Rulings

1. Forward-Looking Preservation Now, Arguments Later

On May 13, 2025, the Court entered an order requiring OpenAI to preserve and segregate output log data that would otherwise be deleted, including data subject to user deletion requests or statutory erasure rights. See id., ECF No. 551. The rationale: once litigation begins, even transient data can be critical to issues like bias and representativeness. The Court stressed that it was too early to weigh proportionality, so preservation would continue until a fuller record emerged.

2. Reconsideration Denied, Preservation Continues

A few days later, when OpenAI sought reconsideration or modification of preservation order, the Court denied the request without prejudice. Id., ECF No. 559. The Court noted that it was premature to decide proportionality and potential sampling bias until additional information was developed.

3. A Negotiated “Sunset” and Privacy Carve-Outs

By October 2025, the parties agreed to wind down the broad preservation obligation. On October 9, 2025, the Court approved a stipulated modification that ended OpenAI’s ongoing preservation duty as of September 26, 2025, limited retention to already-segregated logs, excluded requests originating from the European Economic Area, Switzerland, and the United Kingdom for privacy compliance, and added targeted, domain-based preservation for select accounts listed in an appendix. Id., ECF No. 922.

This evolution — from blanket to targeted, time-limited preservation — shows courts’ willingness to adapt when parties document technical feasibility, privacy conflicts, and litigation need.

Implications For Companies

1. Evidence vs. Privacy: Courts Expect You to Reconcile Both

These rulings show that courts will not accept “privacy law conflicts” as a stand-alone excuse to delete potentially relevant data. Instead, companies must show they can segregate, anonymize, or retain data while maintaining compliance. The OpenAI orders make clear: when evidence may be lost, segregation beats destruction.

2. Proportionality Still Matters

Even as courts push for preservation, they remain attentive to proportionality. While early preservation orders may seem sweeping, judges are open to refining them once the factual record matures. Companies that track the cost, burden, and privacy impact of compliance will be best positioned to negotiate tailored limits.

3. Preservation Is Not Forever

The October 2025 stipulation illustrates how to exit an indefinite obligation: offer targeted cohorts, geographic exclusions, and sunset provisions supported by a concrete record. Courts will listen if you bring data, not just arguments.

A Playbook for In-House Counsel

1. Map Your AI Data Universe

Inventory all AI-related data exhaust: prompts, outputs, embeddings, telemetry, and retention settings. Identify controllers, processors, and jurisdictions.

2. Build “Pause” Controls

Design systems capable of segregating or pausing deletion by user, region, or product line. This technical agility is key when a preservation order issues.

3. Update Litigation Hold Templates for AI

Traditional holds miss ephemeral or system-generated data. Draft holds that instruct teams how to pause automated deletion while complying with privacy statutes.

4. Propose Targeted Solutions

When facing broad discovery demands, offer alternatives: limit by time window, geography, or user cohort. Courts will accept reasonable, well-documented compromises.

5. Build Toward an Off-Ramp

Preservation obligations can sunset — but only if supported by metrics. Track preserved volumes, costs, and privacy burdens to justify targeted, defensible limits.

Conclusion

The OpenAI orders reflect a new judicial mindset: preserve broadly first, negotiate smartly later. AI developers and data-driven businesses should expect similar directives in future litigation. Those that engineer for preservation flexibility, document privacy compliance, and proactively negotiate scope will avoid the steep costs of one-size-fits-all discovery — and may even help set the industry standard for balanced AI litigation governance.

By Alex. W. Karasik, Brian L. Johnsrud, and George J. Schaller

Duane Morris Takeaways:  On October 13, 2025, California Governor Gavin Newsom, issued a written statement declining to sign Senate Bill 7 – called the “No Robo Bosses” Act (the “Act”).  While the Act aimed to restrict when and how employers could use automated decision-making systems and artificial intelligence, Governor Newsom rejected the proposed legislation in terms of the Act’s broad drafting and unfocused notification requirements.  Governor Newsom’s statement reflects an initial rebuttal to a wave of pending AI regulations as states wrestle with suitable AI guidance.  Given the pro-employee tendencies of Governor Newsom and California regulators generally, this outcome is a mild surprise.  Employers nonetheless should expect continued scrutiny of AI regulations before enactment.

This legislative activity surely sets the stage for what many believe is the next wave of class action litigation.

Overview Of SB 7: The “No Robo Bosses” Act

The Act was first introduced in December 2024.  After several amendments, it was passed by the Senate Committee on September 23, 2025 for review and signature by Governor Newsom.  The Act’s key proposals included prohibitions on employers solely using AI to make disciplinary or termination decisions, requiring human input for AI disciplinary or termination decisions, detailed advance notice requirements for use of AI in hiring or employment-related decisions, and post-notice requirements if an employer primarily relied on AI for disciplinary or termination decisions. 

The Act focused on automated-decision making systems (“ADS”) and “employment-related decisions.”  Under the Act, an ADS is defined as “any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that issues simplified output, including a score, classification, or recommendation, that is used to assist or replace human discretionary decision making and materially impacts natural personals.”  With this definition, ADS incorporated a swath of technologies utilized by many employers such as call analytic tools, automated scheduling platforms, keystroke and computer monitoring software, and AI-based training programs.  SB 7 also defined “employment-related decisions” as “any decision by an employer that materially impacts a worker’s wages, benefits, compensation, work hours, work schedule, performance evaluation, hiring, discipline, promotion, termination, job tasks, skill requirements, work responsibilities, assignment of work, access to work training opportunity, productivity requirements, or workplace health or safety.” 

The Act also incorporated various pre-notice and post-notice requirements.  Employers using an ADS system to make employment-related decisions (excluding hiring) would have been required to provide “pre-notice” at least 30-days before deploying an ADS, and 30-day notice to new hires for any ADS use.  Similarly, the Act included “post-notice” provisions regarding post-notices when an employer relied on an ADS to make a discipline, termination, or deactivation decision, and to provide the impacted worker with notice at the time the employment decision is made.  Both notices had requirements for the notice to be written in plain language, directed as a routine worker communication, and provided in an accessible format. 

Violations under the “No Robo Bosses” Act included a proposed civil penalty of $500 per violation, with enforcement authority vested in the Labor Commissioner and public prosecutors of California.  The proposed Act did not include a private right of action.  

Governor Newsom’s Veto Of The Act

Governor Newsom’s veto of the Act centered on concerns of unspecified misuses of ADS technology and unfocused notification requirements.  Governor Newsom did recognize the concerns associated with ADS in employment-making decisions but argued the Act’s “proposed solution fail[ed] to directly address incidents of misuse.”  He also found that the restrictions embedded in the Act were broad and removed “a potentially valuable tool” when ADS systems are properly applied and properly employed.  Governor Newsom’s critique of the Act demonstrates that the Act did not distinguish the benefits of ADS systems compared to risks associated with ADS use cases.   Accordingly, Governor Newsom vetoed SB 7.

Implications Of The Veto

California employers do not have to mitigate their ADS systems yet based on Governor Newsom’s veto of SB 7, but given the Governor’s comments, its possible new legislation will be introduced to narrow the use of ADS systems in employment decisions.  Governor Newsom’s veto of the Act further represents a growing concern among ADS systems and AI technologies legislative policies – namely that broad legislative efforts cannot efficiently or effectively address emerging technologies.  While employers can expect other states may propound ADS and AI legislation in the context of employment decision-making, employers should consider that if the notoriously pro-employee State of California struck down legislation as overly broad and unfocused – it may take some time for other jurisdictions to determine how to finesse the legislative landscape.

Employers should continue to monitor federal developments in this area, as well.  In July 2023, the federal “No Robot Bosses Act,” S.2419, was introduced in the Senate.  While the bill has not been enacted, its provisions include similar limitations on the use of automated systems and would require human oversight before an automated decision is finalized.

By Alex W. Karasik

Duane Morris Takeaways:  Artificial Intelligence has brilliantly transformed society to the point where no industry can fully separate from its impact. But the fruits of this technology must be carefully curated to ensure that its adoption is ethical.  An evolving legislative landscape and billion-dollar class action litigation industry loom large. 

This week, at the University of Notre Dame’s inaugural R.I.S.E. AI Conference in South Bend, Indiana, Partner Alex W. Karasik of the Duane Morris Class Action Defense Group was a panelist at the highly anticipated session, “Challenges And Opportunities For Responsible Adoption Of AI.”  The Conference, which had over 300 attendees from 16 countries, produced excellent dialogues on how cutting-edge technologies can both solve and create problems, including class action litigation.

The Conference covered a wide range of global issues affected by AI.  Some of the topics included AI’s impact on data privacy, information governance, healthcare, education, voting, and its overall impact on Latin America – including discussions about how large language models are developing when machines are trained in non-English languages.  For organizations who deploy this technology, or are thinking about doing so, the Conference was informative in terms of AI’s utility and risk.  The sessions provided valuable insight from a broad range of constituents, including business leaders, world-renowned academic scholars, technology professionals – and a lawyer from Chicago. 

I had the privilege of discussing AI’s integration into the workplace in two areas: (1) proactive implementation; and (2) reactive class action litigation risk. There is no “one-size-fits-all” checklist for organizations to incorporate AI.  But there are several overarching principles that will likely be important factors when establishing an ethical and legally compliant AI framework. These include: (1) creating an AI steering committee with a diverse collection of viewpoints, including Legal, HR, IT, business operations, and other end-users – such as tech-savvy employees – who can collectively opine on the benefits and concerns of AI in the workplace; (2) crafting a robust yet unambiguous policy to ensure that all members of an organization as using AI responsibly and consistently; (3) implementing training programs for both managers and employees on how to equitably implement the AI policy, and understand its interplay with other policies such as EEO; (4) communicating with AI vendors to understand how AI models were trained; and (5) conducting audits before and after implementation to ensure AI use does not result in a disparate impact on certain demographics of applicants or employees.

From a litigation perspective, I discussed the “moving target” of AI laws popping up around the country, which may create compliance challenges.  While most of these laws are guided by the same fundamental principles (i.e. transparency and disclosure when AI is being used in the hiring process), accounting for minor variations may ultimately present compliance challenges for employers with national and international operations.  The class action litigation and EEOC-initiated systemic discrimination litigation will inevitably follow — as the EEOC v. iTutorGroup, Inc., et al., Case No. 1:22-CV-02565 (E.D.N.Y.) settlement (see our blog post) and currently pending Mobley v. Workday, Inc., Case No. 3:23-CV-00770 (N.D. Cal.) class action lawsuit (see our blog post) confirm.

Overall, I was amazed by the amount of business and academic talent at the Conference.  The Conference was an incubator for issue-spotting, brainstorming, and problem-solving.  I am grateful for the opportunity to learn about the statistical impact of AI on organizations – and thankful to my many new PhD friends for sharing explanations of their empirical studies.  Looking forward, I am optimistic that when constituents from all over the world in a variety of professions collaborate together, we will responsibly unlock AI’s greatest potential.

For more information about Duane Morris’s endeavors in the Artificial Intelligence space, please visit our Firm’s AI webpage here.

By Gerald L. Maatman, Jr., Justin Donoho, and George J. Schaller

Duane Morris Takeaways: On October 1, 2025, California’s “Employment Regulations Regarding Automated-Decision Systems” will take effect.  These new AI employment regulations can be accessed here.  The regulations add an “agency” theory under the California Fair Employment and Housing Act (FEHA) and formalize this theory’s applicability to AI tool developers and companies employing AI tools that facilitate human decision making for recruitment, hiring, and promotion of job applicants and employees.  With California’s inclusion of a private right of action under the FEHA, these new AI employment regulations may augur an uptick in AI employment tool class actions brought under the FEHA.  This blog post identifies key provisions of this new law and steps employers and AI tool developers can take to mitigate FEHA class action risk.

Background 

In the widely-watched class action captioned Mobley v. Workday, No. 23-CV-770 (N.D. Cal.), the plaintiff alleges that an AI tool developer’s algorithm-based screening tools discriminated against job applicants on the basis of race, age, and disability in violation of Title VII of the Civil Rights Act of 1964 (“Title VII”), the Age Discrimination in Employment Act of 1967 (“ADEA”), the Americans with Disabilities Act Amendments Act of 2008 (“ADA”), and California’s FEHA.  Last year the U.S. District Court for the Northern District of California denied dismissal of the Title VII, ADEA, and ADA disparate impact claims on the theory that the developer of the algorithm was plausibly alleged to be the employer’s agent, and dismissed the FEHA claim which was brought only under the then-available theory of intentional aiding and abetting (as we previously blogged about here).

In recent years, discrimination stemming from AI employment tools has been addressed by other state and local statutes, including Colorado’s AI Act (CAIA) setting forth developers’ and deployers’ “duty to avoid algorithmic discrimination,” New York City’s law regarding the use of automated employment decision tools, the Illinois AI Video Interview Act, and the 2024 amendment to the Illinois Human Rights Act (IHRA) to regulate the use of AI, with only the last of these laws providing for a private right of action (once it becomes effective January 1, 2026).

Key Provisions Of California’s AI Employment Regulations

California’s AI employment regulations amend and clarify how the FEHA applies to AI employment tools, thus constituting a new development in case theories available to class action plaintiffs regarding alleged harms stemming from AI systems and algorithmic discrimination.  

Employers and AI employment tool developers should take note of key provisions codified by California’s new AI employment regulations, as follows:

• Agency theory.  An “agency” theory is added under the FEHA like the one that allowed the plaintiff in Mobley v. Workday to proceed past a motion to dismiss on his federal claims, whereby an AI tool developer may face litigation risk for developing algorithms that result in a disparate impact when the tool is used by an employer.  While Mobley v. Workday continues to proceed in the trial court, no appellate authority has yet had occasion to address the “agency” theories being litigated in that case under federal antidiscrimination statutes.  However, with the California AI employment regulations taking effect October 1, 2025, that theory is now expressly codified under the FEHA.  2 Cal. Code Regs § 11008(a).
• Proxies for discrimination.  The regulations clarify that it is unlawful to use an employment tool algorithm that discriminates by using a “proxy,” which the regulations define as a “characteristic or category closely correlated with a basis protected by the Act.”  Id. §§ 11008(a), 11009(f).  While the regulations do not explicitly identify any proxies, proxies that have been identified in literature by the EEOC’s former Chief Analyst include zip code (this proxy is also codified in the IHRA), first name, alma mater, credit history, and participation in hobbies or extracurricular activities.
• Anti-bias testing.  The regulations state that relevant to a claim of employment discrimination or an available defense are “anti-bias testing or similar proactive efforts to avoid unlawful discrimination, including the quality, efficacy, recency, and scope of such efforts, the results of such testing or other effort, and the response to the results.”  Id. § 11020(b).  Thus, for example, adoption of the NIST’s AI risk management framework, itself codified as a defense under the CAIA, could be a factor to consider as a defense under the FEHA.  Many other factors are pertinent with respect to anti-bias testing, including auditing, tuning, and the use of various interpretability methods and fairness metrics, discussed in our prior blog entry and article on this subject (here).
• Data retention.  The regulations provide that employers, employment agencies, labor organizations, and apprenticeship training programs must maintain employment records, including automated-decision data, for a minimum of four years.  Id. § 11013(c).

Implications For Employers

California’s AI employment regulations increase employers’ and AI tool developers’ risks of facing class action lawsuits similar to Mobley v Workday and/or alleging discrimination under the FEHA.  However, developers and employers have several tools at their disposal to mitigate AI employment tool class action risk.  One is to ensure that AI employment tools comply with the FEHA provisions discussed above and with other antidiscrimination statutes.  Others include adding or updating arbitration agreements to mitigate the risks of mass arbitration; collaborating with IT, cybersecurity, and risk/compliance departments and outside advisors to identify and manage AI risks; and updating notices to third parties and vendor agreements.

By Justin Donoho

Duane Morris Takeaway: Available now is the recent article in the Journal of Robotics, Artificial Intelligence & Law by Justin Donoho entitled “Five Human Best Practices to Mitigate the Risk of AI Hiring Tool Noncompliance with Antidiscrimination Statutes.”  The article is available here and is a must-read for corporate counsel involved with development or deployment of AI hiring tools.

While artificial intelligence (AI) hiring tools can improve efficiencies in human resource functions, such as candidate sourcing, resume screening, interviewing, and background checks, AI has not replaced the need for humans to ensure that AI-assisted human resources (HR) practices comply with a wide range of antidiscrimination laws such as Title VII of the Civil Rights Act of 1964 (Title VII), the Americans with Disabilities Act (ADA), the Age Discrimination in Employment Act (ADEA), the sections of Colorado’s AI Act setting forth developers’ and deployers’ “duty to avoid algorithmic discrimination” (CAI), New York City’s law regarding the use of automated employment decision tools (NYC’s AI Law), the Illinois AI Video Act (IAIVA), and the 2024 amendment to the Illinois Human Rights Act to regulate the use of AI (IHRA).  This article identifies human best practices to mitigate the risk of companies’ AI hiring tools violating the foregoing statutes, according to the statutes, EEOC regulations, and scholarly sources authored by EEOC personnel and leading data scientists.

Implications For Corporations

AI hiring tools designed to comply with antidiscrimination statutes will comply.  Moreover, by eliminating some human decision-making and replacing it with carefully designed algorithms, AI holds the potential to substantially reduce the kind of bias that has been unlawful in the United States since the civil rights movement of the mid-twentieth century.  This article identifies human best practices to assist with such compliance and, relatedly, such potential substantial reduction of bias.