New Data Shows Sharp Rise in AI-Related Litigation Concentrated Among Few Defendants
Generative artificial intelligence has moved from experimental tool to embedded infrastructure across a wide swath of American business in a remarkably short window, and the federal courts are beginning to register that shift in a measurable way. A new analysis from DOAR, a national trial consulting firm, examined 168 AI-related cases filed in federal district courts and found both a sharp acceleration in filings and a striking pattern in who is being sued: a small number of defendants account for a disproportionate share of the litigation.
For companies that build, license, or deploy AI tools, and for the litigators who represent them, the data offers one of the first systematic pictures of where this emerging category of disputes is actually landing, and what that concentration might mean for exposure going forward.1
A Sharp and Recent Surge in Filings
DOAR’s methodology was deliberately narrow, and the exact shape of the screen matters for reading everything that follows. The figures come from a targeted, manual review of cases involving companies clearly engaged in artificial intelligence, focused on disputes where AI technology is central to the matter and excluding cases where AI is incidental or not clearly implicated.2 Two limits follow from that design. The dataset is built around AI companies as defendants, so claims arising from AI tools deployed by ordinary employers, lenders, retailers, or health plans generally fall outside it even when the technology is squarely at issue. And because peripheral mentions of AI are excluded, disputes where the technology is present but is not the gravamen of the complaint drop out as well. The counts below are best understood as a floor rather than a full census, which makes the growth curve they describe more, not less, notable.
The case count by year tells the story plainly. AI-related federal filings were minimal in the early period covered by the dataset, with seven cases in 2022 and nineteen in 2023. Filings ticked up modestly to twenty-two in 2024. Then, in 2025, volume surged to ninety-four cases in a single year, more than half of all filings in the entire multi-year dataset. Filings in 2026 remained elevated through the report’s April publication date, with 26 cases recorded so far that year; DOAR noted the pace suggested elevated activity was likely to continue even if 2026 does not ultimately match 2025’s spike.3
That inflection point roughly tracks the broader commercial adoption curve for generative AI tools, from chatbots and coding assistants to AI-driven hiring, underwriting, and content platforms. As those tools moved from pilot programs into production use across industries, the range of plaintiffs, injuries, and legal theories available to challenge them expanded correspondingly.
Where the Cases Are Being Filed
Geography matters in this dataset. AI litigation is not evenly distributed across the federal judicial districts; it is concentrated in a handful of courts. The Northern District of California leads by a wide margin, with 53 of the 168 cases, reflecting its proximity to the major AI developers headquartered in Silicon Valley. The Southern District of New York follows with 23 cases, consistent with its traditional role as a hub for copyright, media, and financial litigation involving technology companies. The District of Columbia and the Western District of Texas each recorded nine cases, with the Central District of California and the Eastern District of Texas also registering meaningful activity.4
That geographic concentration has practical consequences for litigation strategy. Judges in the Northern District of California and the Southern District of New York are accumulating substantial experience with AI-specific evidentiary and procedural issues, from expert discovery on training data and model architecture to the treatment of AI outputs as potential infringing or defamatory content. Parties in AI disputes filed elsewhere are increasingly likely to find those courts’ rulings cited as persuasive authority, even outside a formal multidistrict litigation structure.
A Small Number of Defendants Carry an Outsized Share
The most consequential finding in the DOAR analysis may be the degree to which AI litigation is concentrated among a small group of repeat defendants. OpenAI alone was named in 71 of the 168 cases, more than 40 percent of all filings in the dataset, a reflection of its central and highly visible role in the generative AI ecosystem and its position as a natural target for plaintiffs testing novel legal theories.5
The concentration of claims against a limited set of frontier AI developers is visible in how those cases have been managed procedurally. In April 2025, the Judicial Panel on Multidistrict Litigation centralized a dozen related copyright cases against OpenAI and Microsoft in the Southern District of New York, consolidating disputes including Authors Guild v. OpenAI, several other author and publisher suits, and The New York Times Company’s litigation against Microsoft and OpenAI.
That proceeding, now captioned In re: OpenAI, Inc., Copyright Infringement Litigation, has already produced substantive rulings; in October 2025, the presiding judge denied OpenAI’s motion to dismiss claims that ChatGPT outputs themselves could constitute infringing reproductions of copyrighted works, allowing that theory to proceed alongside the underlying training-data claims.6
OpenAI is not the only company facing repeat litigation from a single legal theory. Facial-recognition company Clearview AI, for example, resolved a long-running Illinois biometric privacy class action covering claims that it scraped billions of facial images from the internet without consent, agreeing to a settlement valued at over fifty million dollars, structured in part as an equity stake rather than a cash payout.7
That kind of concentrated, repeat exposure is exactly what DOAR’s data captures at a category-wide level: a small set of companies operating at the center of the AI ecosystem are absorbing a disproportionate share of both filings and the reputational and financial risk that comes with defending them.8
The Legal Theories Driving the Numbers
A caveat before the numbers get pushed further than they go: DOAR did not break its dataset out by cause of action, and the report publishes no claim-type tally. What DOAR does observe is that plaintiffs are taking a broad, exploratory approach, testing multiple overlapping claims to see which legal pathways gain traction, and that the current wave is best understood as an effort to stretch existing doctrines — particularly in intellectual property, data usage, privacy, and liability — to fit a technology those doctrines were not designed for.9
My own reading of the reported decisions and dockets is consistent with that description, though it is an impression rather than a count. Copyright and other intellectual property claims are the most visible cluster, driven by the training-data and output-reproduction disputes described above. Privacy and biometric claims, often brought under state statutes like Illinois’s Biometric Information Privacy Act, form a second recognizable group. A newer set of cases advances product liability and wrongful-death theories arising from harm allegedly caused by AI chatbots, particularly claims involving vulnerable users.
Two categories that in-house counsel ask about most often — employment claims challenging AI-driven hiring, screening, and evaluation tools, and consumer protection claims alleging deceptive or unfair AI-related business practices — are largely invisible in this dataset, and for a structural reason. Those suits typically name the employer, lender, or merchant that deployed the tool, not a company “clearly engaged in artificial intelligence,” so DOAR’s screen filters them out. Their thin presence in the case counts says something about how the data was built, not about how much risk is out there.
AI-related exposure is also surfacing in places this dataset does not reach at all. Securities class actions tied to AI representations and disclosures were, in Stanford professor Joseph Grundfest’s description of first-half 2026 filing data, “a modest share of total filings but an outsized share of alleged investor losses.” Both halves of that sentence matter. The filing counts are not yet dramatic; the dollars behind them are. And the point underscores that AI litigation risk extends well past product-liability and IP theories into the disclosure and governance obligations that come with marketing a company’s AI capabilities to investors.10
How Jurors Are Likely to View These Cases
The litigation-trends analysis says nothing about how these disputes will play to a jury. DOAR’s jury consultants have taken up that question separately, in work not tied to the April report, and it is an increasingly relevant one as the current wave of filings grinds through motion practice and discovery toward actual jury pools.
In a question-and-answer piece published in May 2026, DOAR Director Ellen Brickman, Ph.D., a jury consultant, offered her impressions of how jurors may respond to chatbot cases. These are professional judgments rather than survey or mock-trial findings, and worth reading as such. Her expectations: publicity around chatbot-related harm will most sharply move jurors who are already least comfortable with AI; the emotional attachments users form to chatbots will be genuinely novel territory for a jury; a one-line disclaimer warning users not to rely on chatbot advice is unlikely to carry much weight where the user is a teenager or otherwise vulnerable, particularly because jurors may see products engineered to hold a user’s attention as inducing the very reliance the disclaimer purports to disclaim; and the facts most likely to matter are user vulnerability and the effort a company did or did not make to build in safeguards.11
The firm’s related public-opinion research adds harder numbers to that impression. A national survey of 1,010 jury-eligible respondents, released in March 2026, found that roughly two-thirds believe companies that own AI chatbots should bear responsibility when a user’s suicide follows interactions with the chatbot, and that more than three-quarters supported safety guardrails such as barring chatbots from discussing suicide methods or escalating concerning conversations to human reviewers. The same survey found that respondents who use AI regularly were less likely to assign blame to the company, a wrinkle worth noting for voir dire.12
A separate and considerably broader study of civil liability decision-making, drawing on more than 2,000 jury-eligible participants across 64 civil focus groups and mock trials conducted between 2020 and 2025, points in a direction that should reassure defense counsel more than it worries them. The study’s central findings were that jurors’ pre-deliberation opinions remained highly predictive of their ultimate verdict preferences, and that perceived case strength was the strongest predictor of liability outcomes: as jurors perceived the defense case to be stronger, the likelihood of a liability finding declined substantially, regardless of demographic differences. Deliberation still moved a meaningful minority — nearly one in four defense-leaning jurors ultimately found liability after group discussion — so initial attitudes are not immovable. The study examined civil litigation generally rather than AI cases, but the practical lesson translates: jury selection matters because early leanings largely hold, and it is no substitute for building a case that jurors perceive as strong.13
Taken together, these findings suggest that AI defendants should not assume that public familiarity with, or even enthusiasm for, AI tools will translate into juror sympathy at trial. A substantial share of the public is already predisposed to assign responsibility to the company behind a chatbot when a user is harmed, and both the survey data and the consultants’ observations point to alleged harm involving vulnerable users, and to decision-making a company cannot explain in plain terms, as the settings where that predisposition is likely to be strongest. Neither survey tested how jurors would apply a particular legal standard of care, however, so this research is better read as a signal about juror predisposition than as a prediction about how negligence or product-liability elements will actually be resolved at trial.
What This Means for Companies Deploying AI
The concentration data does not mean that litigation risk is limited to the handful of frontier developers named most often. It means that risk is currently concentrated there, and companies further down the AI supply chain, from enterprise customers to smaller vendors building on top of major models, should not read the numbers as license for complacency.
First, companies should inventory where AI tools actually touch consumer-facing decisions, employment decisions, or copyrighted or personal data, since those are the precise intersections generating claims today. A company that licenses a third-party model for internal use faces a different risk profile than one that deploys a customer-facing chatbot or an AI-driven hiring tool, and that distinction should drive where compliance and legal review resources are focused.
Second, businesses should revisit vendor contracts and indemnification provisions with AI providers now, while the market for those terms is still developing, rather than after a claim has already been filed. Given how concentrated the current wave of litigation is among a small number of developers, downstream users have real leverage to negotiate meaningful indemnification and cooperation obligations, and that leverage will likely erode as the market matures and standard-form terms harden.
Third, in-house teams should coordinate with insurance brokers to confirm whether existing technology errors-and-omissions, cyber, or general liability coverage actually responds to AI-related claims, including the biometric privacy, employment discrimination, and product liability theories now appearing in AI-related litigation, rather than assuming legacy policy language was drafted with these scenarios in mind.
Fourth, companies using AI tools in hiring, lending, or other consequential decisions should document the human oversight built into those processes, since both the litigation data and the juror-perception research point toward opaque or fully automated decision-making as a particular flashpoint for liability theories and juror skepticism alike.
Fifth, companies operating or deploying consumer-facing chatbots, particularly those that may interact with minors or other vulnerable users, should treat the wrongful-death and product-liability theories emerging in this space as a present risk rather than a speculative one, and should build escalation and human-intervention protocols accordingly.
Looking Ahead
The DOAR data captures an inflection point rather than a settled landscape. With 2025 filings alone exceeding the combined total of the three prior years, and with 2026 tracking toward another elevated year, AI-related litigation is unlikely to plateau in the near term. The concentration of claims among a small number of frontier developers may persist as those companies continue to expand their products and their user bases, but the underlying legal theories — copyright, privacy, employment, product liability, and securities disclosure among them — are precisely the kind of claims that migrate downstream to enterprise customers and smaller vendors as adoption spreads. Companies that treat the current data as an early warning, rather than someone else’s problem, will be better positioned when that migration accelerates.