There is a quiet assumption baked into every AI search product: that the open web is raw material, free for the taking, and that pointing users to a tidy synthesized answer is a favor to everyone. In 2026, that assumption is being tested in court — and the bill is coming due.
The flashpoint is AI search copyright, the legal and ethical question of whether an AI engine can ingest a publisher's journalism, summarize it, and serve that summary to users who never click through. More than 40 copyright cases are now active between AI companies and rights holders, and the answers emerging from settlements and courtrooms will redraw the economics of the web.
The Case Against Perplexity
No company embodies the fight better than Perplexity. In December 2025, The New York Times sued the startup in the Southern District of New York, accusing it of unlawfully scraping Times stories, videos, and podcasts to generate answers — and producing outputs "identical or substantially similar to" the original reporting.
The complaint makes two distinct arguments worth separating:
- Substitution. Perplexity's summaries allegedly compete directly with the Times, giving users the substance of an article without the visit that funds it. That undercuts both subscription and advertising revenue.
- Misattribution. The Times says Perplexity hallucinates false information and attaches the paper's name to it — a brand-damage claim distinct from copying.
The Times is not alone. CNN filed suit alleging Perplexity copied and redistributed roughly 17,000 of its stories, including material behind paywalls. Dow Jones and the New York Post brought their own infringement claims, and Reddit sued over large-scale scraping of its platform. A single product is now fighting a multi-front war.
The Paywall Problem
The ethical core of the dispute isn't summarization in the abstract — it's how the content gets in.
Researchers found that Perplexity's Comet browser could retrieve the full text of subscriber-only articles. The mechanism is almost banal: many paywalls are client-side, meaning the full text already loads into the page and is merely hidden behind a translucent overlay. A human sees a wall. An AI agent reads straight through it, because the words were never actually withheld — just visually obscured.
When a paywall is a curtain rather than a vault, an agent that ignores the curtain isn't "breaking in." It's exposing that the lock was decorative all along.
That nuance matters ethically. Publishers built flimsy paywalls because they were optimizing for search-engine crawlers and casual readers, not autonomous agents. The technology outran the architecture, and now the gap is a courtroom exhibit.
What the Anthropic Settlement Signals
If publishers want a preview of leverage, they got one in Bartz v. Anthropic. In late August 2025, Anthropic agreed to pay $1.5 billion to settle a class action over books it had downloaded from pirate sites — the largest copyright settlement in U.S. history, working out to roughly $3,000 per book across an estimated 500,000 titles.
Read the fine print, though. The settlement releases Anthropic only for past acquisition of identified pirated works before August 25, 2025. It creates no ongoing license, and it explicitly does not constrain training on lawfully acquired material. In other words, it punished how the data was obtained, not the act of training itself.
That distinction is the whole game. Courts so far seem far more hostile to piracy and scraping than to AI training as a concept. For AI search, where the alleged harm is verbatim reproduction and revenue substitution rather than abstract model-training, that posture could cut sharply against the platforms.
The Industry's Two Responses
Faced with mounting suits, AI companies are splitting into two camps.
The first is litigate and argue fair use — bet that transformation and public benefit will shield them, and absorb settlements as a cost of doing business.
The second is pay the toll. In early January 2026, Perplexity launched Comet Plus, a publisher program that shares revenue when a publisher's content is cited in an answer, and routes users to paywalled content from partners such as CNN, Fortune, and The Washington Post. The timing — arriving amid a wave of lawsuits — tells you what it is: a peace offering and a hedge.
Both responses concede the same underlying point. The "open web is free fuel" era is ending. The only real debate now is the price and who sets it.
Why This Is an Ethics Story, Not Just a Legal One
It is tempting to file this under intellectual-property law and move on. That misses the stakes.
Journalism is expensive to produce and cheap to summarize. If AI search captures the value of reporting while starving the institutions that do the reporting, the result isn't just unfair to publishers — it degrades the information commons that the AI systems themselves depend on. A model trained on a web that no longer funds original journalism is a model slowly poisoning its own well.
There is a counter-argument worth stating fairly: AI answers can expand access, surface obscure sources, and serve users that publishers' own paywalls shut out. Synthesis is a genuine good, and not every summary is a substitute for the original. The honest position is that both things are true at once — and that a sustainable settlement has to preserve the incentive to create the work in the first place.
The Bottom Line
The AI search copyright fight is not really about whether machines can read. It's about who pays for the reading. The Anthropic settlement showed that scraping and piracy carry real, ten-figure consequences, and the pileup of suits against Perplexity shows that publishers have stopped waiting for permission to push back. Revenue-sharing programs like Comet Plus are a tacit admission that the content was never free. Expect more deals, more lawsuits, and — eventually — a market where access to journalism is licensed, not lifted. The companies that figure out the price first will be the ones still standing when the courts finish the math.
