This year marks three years since The New York Times sued OpenAI and Microsoft for copyright infringement. The outcome could become a milestone in clarifying whether AI vendors may train models on large volumes of creative content without the creators’ permission — but the case is still pending, and that delay says as much about the state of AI law as any ruling could. The Times litigation is a useful backdrop for what 2026 looks like in the conflict between creators and AI vendors: slow landmark cases, accelerating settlements, and a licensing market forming in the gaps.
The Case That Frames Everything Else
In its lawsuit filed on December 27, 2023, the Times accused OpenAI and its principal backer Microsoft of using millions of copyrighted articles without permission to train generative AI models, arguing that the resulting tools compete directly with the newspaper’s business. OpenAI has countered that its use of publicly available material is transformative and that the publisher’s position would stifle innovation.
More than two years after the Times filed — and three years after Getty Images sued Stability AI over image training data — neither case has reached a definitive conclusion, and analysts consider it unlikely they will finish in 2026. What has changed is everything around them: creators and model makers increasingly meet at the negotiating table instead of the courtroom.
Fair Use and the Transformation Question
The core legal question remains fair use — the doctrine that permits use of copyrighted material without permission for purposes such as news reporting, research, and other uses serving the public interest — and, within it, transformation: whether AI training turns source material into something new in purpose and character. Analysts note that courts have so far been more receptive to fair-use arguments where AI systems are general-purpose tools rather than direct competitors to the plaintiff’s business.
Courts began engaging seriously with the question in 2025. In June of that year, U.S. District Judge William Alsup ruled that, given the transformative character of large language model training, fair use was a plausible defense in a case brought by a group of authors against Anthropic — but allowed the case to proceed to trial over Anthropic’s use of pirated books. In September 2025, Anthropic agreed to pay $1.5 billion to authors, an amount widely described as the largest copyright settlement in American history.
The lesson analysts draw: fair use will remain the central argument, but in 2026 courts are focusing increasingly on how training data was collected — whether it was pirated, and whether its acquisition violated contractual terms — rather than on the abstract legality of training itself.
More Settlements Coming
The Anthropic agreement is expected to accelerate a settlement wave. As Kashyap Kompella, CEO and founder of RPA2AI Research, put it in the original analysis published by AI Business, a single large settlement resets expectations across the plaintiff bar and the litigation-finance ecosystem, increasing pressure to resolve cases once core facts are established. Publishers and authors, for their part, now use the Anthropic figure as a benchmark when deciding whether to push for trial.
The deeper logic favoring settlement is precedent risk. Michael McCready, owner of Chicago-based McCready Law, notes that a trial is a gamble for both sides: a loss sets a binding precedent not just for the parties but for everyone aligned with them. If plaintiffs lose, creators get nothing; if a major vendor loses at trial, the damages exposure could dwarf any negotiated figure. Michael Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago, considers The New York Times a likely settlement candidate in 2026, given the sheer volume of journalistic content allegedly involved and the incentive the Anthropic settlement created. Reputational exposure adds pressure — large AI companies face not only legal risk but the stigma of being publicly accused of building products on uncompensated creative work. What any vendor can actually pay, Kompella cautions, still depends on its finances — a constraint that separates the frontier labs from smaller model makers.
More Licensing Deals — but No Industry Standard
Alongside settlements, licensing agreements are multiplying. The New York Times itself struck a licensing deal with Amazon in May 2025, reportedly worth $20 million to $25 million. Google’s agreement with Reddit allows user-generated content to feed its Gemini models. Perplexity AI runs a publisher program — with partners including the Los Angeles Times — that shares revenue when its AI search engine draws on partner content in generated answers.
There is an irony in the pattern: several companies that accused AI vendors of infringement have since become their licensing partners, converting legal leverage into recurring revenue. But analysts see no collective, industry-wide standard emerging in 2026 — no equivalent of the blanket licensing that governs music. The closest historical parallel is instructive: in 2001, Napster agreed to pay $26 million to settle music-sharing lawsuits; the deal collapsed when Napster entered bankruptcy and a judge blocked its acquisition by Bertelsmann, yet the episode laid the groundwork for the compensation model that now underpins licensed music streaming.
Bennett doubts AI licensing will reach that scale of coordination soon, because the training corpus is too heterogeneous: journalism, fiction, code, images, and music all carry different levels of protection and different bargaining power. Kompella sees the plausible consensus elsewhere — enforceable dataset transparency, scalable licensing for high-value corpora such as publisher archives, music catalogs, and stock libraries, and output-side guardrails like provenance tooling and watermarking.
Beyond Copyright: Bias Cases and Bigger Issues
Bennett also expects intellectual-property litigation to recede somewhat in relative prominence as attention shifts to AI’s effects on employment, education, and energy consumption. A second front is already active: algorithmic discrimination. In Mobley v. Workday, plaintiff Derek Mobley alleges that an AI-based hiring screen harmed his job applications; in Massachusetts, Mary Louis and Monica Douglas sued tenant-screening firm SafeRent Solutions over an algorithm they alleged discriminated against Black renters — a case SafeRent settled for $2.275 million. Kompella observes that bias cases tend to resolve through operational remedies — audits, monitoring, usage limits — rather than categorical bans.
Courts Are Doing Regulators’ Work
James Cooper, a professor at California Western School of Law, argues that whatever the case type, clarity has to come from somewhere — and that regulators, not judges, should be providing it. For now, individual courts and local jurisdictions are effectively setting AI policy case by case, while binding legislative guidance remains absent. Few expect that to change quickly: McCready is blunt that congressional consensus is unlikely, given the competing interests at stake.
Limitations and What to Watch
Predictions about litigation are exactly that — the expert views summarized here are informed forecasts, not outcomes, and settlement figures reported for private deals (such as the Amazon–Times arrangement) rest on press reporting rather than disclosed terms. Case status changes quickly; readers checking the current state of any matter should consult a live tracker such as BakerHostetler’s AI case tracker. The three developments most worth watching in 2026: whether the Times–OpenAI case settles or heads toward trial, whether any court issues a merits ruling that squarely addresses training-data fair use, and whether licensing consolidates into standard terms or remains a patchwork of bilateral deals. For how these legal risks intersect with business adoption, see this related piece on the falling cost of AI agents for small businesses.
