Anthropic Pays for News for the First Time as It Licenses AP Election Data for Claude

Anthropic, the artificial intelligence company behind the Claude chatbot, has paid a news publisher for its content for the first time. The company struck a licensing deal with the Associated Press just before the United States midterm elections, according to Adweek, ending years in which it had refused to negotiate with publishers at all — and reversing a position it had defended on the grounds of fair use.

The deal, reported by Adweek on 5 November and confirmed by further trade coverage the same day, allowed Claude to show voters live, attributed election results on election night. Anyone who opened Claude on 3 November and asked for the latest polling data would have been given up-to-the-minute information provided by and attributed to the Associated Press, according to Adweek’s report, which cited two people familiar with the matter. Neither Anthropic nor the AP commented on the record, and neither side disclosed the financial terms of the arrangement, its duration, or whether it extends beyond election coverage.

That silence is itself telling. The agreement is narrow in scope and was struck quietly, yet it marks a strategic reversal for one of the most closely watched companies in the world. Anthropic had, until this week, never signed a content licensing deal with any digital publisher, according to Adweek — a stance that set it apart from its two main rivals, OpenAI and Google, both of which have spent years building licensing relationships with the news industry. OpenAI has now negotiated more than a dozen content licensing deals with major publishers, including Axel Springer, News Corp and The Atlantic, according to Adweek, while Google said in October that more than 400 publications were taking part in its licensing programme.

The timing is not accidental. Anthropic filed the paperwork for its stock market debut on 26 October, disclosing third-quarter revenue of about $17 billion, and is expected to begin marketing its shares to investors in the coming weeks. A company about to ask the public markets to value it at as much as $2 trillion can no longer afford to have the most consequential question in its industry — whether its business model depends on taking other people’s work without paying — hanging over it unanswered. The AP deal does not settle that question. But it is the first concrete evidence that Anthropic is prepared to pay for at least some of the information its products depend on, and it arrives in the same week that the first cheques from its $1.5 billion copyright settlement with book authors are due to reach claimants.

The deal, and what Claude users saw on election night

The arrangement was struck, in Adweek’s words, “just before the midterm elections”, and its effects were visible immediately. Anthropic had announced earlier in the autumn that Claude would begin showing “up-to-date results on election night and in the days that follow”, according to Dealroom’s account of the deal, and the AP agreement gave it a credible source for that live data just as voters went to the polls.

Until then, the chatbot had a conspicuous blind spot. Because Anthropic had refused to strike licensing deals with major news publishers, it lacked access to the kind of credible, timely reporting that consumers turn to at key moments such as an election, Adweek’s report explained. The publishers that matter most on a night like that — the wire services, the national newspapers, the broadcasters — block Anthropic’s web crawlers, and without their data Claude could only draw on what AI researcher Deana Burke calls the “unlocked web”: sites that do not block AI crawlers, which, as Adweek put it, “might serve as sufficient sources of information for low-stakes queries, but they cannot tell a user which politician won a Senate seat 10 minutes ago.”

The midterm elections were the worst possible moment for that weakness to be exposed. The 3 November vote was one of the most consequential elections of the decade, handing Democrats control of both chambers of Congress and setting up two years of divided government, and it generated one of the biggest surges in search and information queries of the year. A chatbot that could not tell its users who had won, accurately and promptly, on election night would have looked close to useless as a source of real-time information — and Anthropic, which is trying to position Claude as the serious, trustworthy alternative to its rivals, could not afford that impression. The AP deal solved the problem at the last minute, and it solved it with the one partner whose data carries the most authority in American elections: the AP’s vote counts and race calls are the standard against which every other outlet measures its own results.

How far the deal goes beyond that night remains deliberately unclear. Adweek noted that one deal “does not necessarily signal a wholesale reversal” of Anthropic’s licensing strategy, and reported that the company is likely to use the data from the experiment to decide whether to go further down this path. The AP, for its part, is an exceptional case among publishers, and the deal may tell us less about a change of heart at Anthropic than about the specific arithmetic of this one partnership.

Why the AP was the ideal first partner

The Associated Press is not a typical publisher, and that is precisely why it makes sense as Anthropic’s first licensing partner. Founded in 1846, the AP is a non-profit news cooperative owned by its member newspapers and broadcasters, and it generates the vast majority of its revenue through licensing and wire deals with other publishers. Much of its business, as Adweek observed, is already structured around striking exactly these kinds of tie-ups. Where a newspaper or magazine must weigh whether licensing its journalism to an AI company will cannibalise its own readership and advertising, the AP’s business model is built on selling its reporting to others in the first place.

The AP was also the first publisher with whom both Google and OpenAI signed licensing agreements, according to Adweek, which makes it the veteran of this particular negotiation. It knows what its data is worth, it knows how to structure a deal, and it knows how to protect the attribution that is its most valuable asset. For Anthropic, that experience lowered the cost of a first step: the company could buy the election data it needed from a partner that was already set up to sell it, without having to negotiate the thornier questions — training data, model weights, the use of archives — that a deal with a newspaper would raise.

The AP also has a second advantage, which Adweek identified: it is a canonically non-partisan outlet. It positions itself, in Adweek’s phrase, “more of a service, like Getty Images for information, than a standard publisher, which often has an editorial slant one way or the other”. For an AI company that wants to test licensing without being accused of favouring one political viewpoint, a partner whose product is treated as a neutral utility is the safest possible choice. And the AP is now the only publisher with paid licensing agreements across all three of the dominant generative AI platforms — OpenAI, Google and Anthropic — a position that strengthens its hand in every future negotiation it conducts.

None of this makes the deal trivial. It is, after all, the first time Anthropic has paid for news, and the first crack in a position the company had maintained for years. But it does explain why the crack appeared here first, and why it would be a mistake to read the AP deal as proof that Anthropic has abandoned its fair-use argument altogether. A narrow, election-specific licence from a wire service is a long way from a comprehensive licensing programme covering training data, and Anthropic has given no indication that it intends to take the second step.

Why Anthropic held out — and why it gave in now

Anthropic’s reluctance to pay for content had become a subject of open intrigue inside the media industry. In August, Adweek’s Mark Stenberg explored the situation in a piece whose central finding was that the company genuinely believed its scraping and use of publisher content was legally permissible under the concept of fair use. Digital media executives told him at the time that Anthropic had not moved to enter into any licensing partnerships, nor had it indicated any plans to do so in the future. The position was consistent with the company’s public identity: Anthropic presents itself as the safety-conscious, principle-driven AI lab, and paying for content it believed it was entitled to use for free would have looked like a concession it did not need to make.

What changed was not the law. It was the leverage. Publishers had been blocking Anthropic’s crawlers, and those of other answer engines, in a coordinated effort to force negotiations — a tactic Stenberg described as a digital embargo, with news outlets withholding their content in the hope of bringing AI companies to the bargaining table. In this limited instance, the strategy worked. Anthropic discovered that an information index without access to up-to-date data is, in Adweek’s words, “dramatically limited in its utility”, and that the limitation was most acute at precisely the moment when users most needed the product to work.

The economics of the standoff had also shifted underneath Anthropic. In its early years, the company could treat publisher content as a training input — material to be absorbed into a model and never shown to a user. But Claude has evolved from a research tool into a consumer product that answers questions in real time, and real-time answers require real-time data. A model trained on last year’s news cannot tell a user who won last night’s election, however good its training data was. The AP deal is, in that sense, the product of a technical reality as much as a commercial one: the more useful Claude becomes as a live information service, the more it depends on the very publishers it had refused to pay.

The timing, in the middle of Anthropic’s IPO window, supplied the final push. A company preparing to go public is examined by investors, lawyers and regulators in a way that a private company is not, and the question of whether its core product rests on legally contested foundations is exactly the kind of question that examination raises. Paying for election data does not resolve the copyright litigation that still surrounds Anthropic, but it does give the company a concrete example to point to when investors ask whether it can operate in a world where content has a price. It also signals to publishers that the door to negotiation is open, which may matter more than the deal itself: the AP agreement establishes that Anthropic will pay, under the right circumstances, and every other publisher will now test what those circumstances are.

The $1.5 billion shadow over the deal

No account of Anthropic’s change of course is complete without the settlement that now defines its relationship with the creative industries. In July 2026, a federal judge in California gave final approval to a $1.5 billion class-action settlement resolving claims that Anthropic had downloaded millions of copyrighted books from unauthorised online libraries and used them to help train its Claude models. The settlement, the largest known copyright recovery in American history, covers about 482,000 works, with each eligible work worth roughly $3,000, and the first payments to authors and publishers are due on or before 15 November 2026 — within days of the AP deal being disclosed.

The case, Bartz v. Anthropic, drew a line that matters for everything that follows. In June 2025, Judge William Alsup ruled that training an AI model on books Anthropic had lawfully acquired was fair use, but that building a library from pirated copies downloaded from the shadow libraries LibGen and PiLiMi was not. Anthropic resolved the claims without admitting wrongdoing, and agreed to destroy the pirated files. The distinction the court drew — between the legality of training and the illegality of how the training data was obtained — is now the framework within which every AI company’s data practices are judged, and it is a framework that makes the AP deal look less like a courtesy and more like a necessity.

Anthropic still faces other copyright litigation. Music publishers sued the company in 2023 over the use of song lyrics, and further cases from news organisations remain active, according to trade coverage of the AP deal. The New York Times is separately suing OpenAI and Microsoft over copyright, in a case that has become the industry’s defining legal battle. Against that backdrop, a licensing deal with the AP serves a second purpose beyond the election data itself: it demonstrates, to courts, to investors and to the publishers still suing, that Anthropic is willing to pay for content when the content is current, valuable and clearly owned. That is a very different proposition from the book case, where the works were acquired from pirate libraries, but the direction of travel is unmistakable. The company that once argued it did not need to pay anyone is now paying, and the question is no longer whether it will pay, but how much, to whom, and for what.

The settlement’s own afterlife illustrates how messy that question has become. According to reporting by TechCrunch in September, authors who expected straightforward payments have discovered that publishers and literary agents have filed competing claims on their books, with disputes over who owns the rights to works that have gone out of print or reverted to their authors. Claims were submitted for about 93 per cent of the works on the settlement’s list, and the administrator has had to sort out which claimants are entitled to what share. Anthropic itself has no further role in how the money is divided — it has already paid the $1.5 billion into a settlement fund — but the disputes are a preview of the complexity that awaits any AI company that tries to license content at scale. Paying for data is not simply a matter of writing a cheque; it requires knowing who owns what, and in the creative industries that question is rarely simple.

An IPO-season decision

The AP deal also has to be read as a decision taken in the middle of one of the most closely watched stock market debuts in years. Anthropic filed its public prospectus with the Securities and Exchange Commission on 26 October, and filed an amended version on 2 November, disclosing third-quarter revenue of about $17 billion — up 47 per cent on the previous quarter and more than twelve times the figure a year earlier — alongside purchase and lease commitments of $895.4 billion, almost all of it for computing power and infrastructure, running through 2045. The company plans to list its Class A shares on the Nasdaq under the ticker ANT, and reports have put its possible valuation at up to $2 trillion, with a roadshow to investors expected in the coming weeks and the offering set before the American Thanksgiving holiday, according to Bloomberg.

Those numbers explain both the confidence and the caution in Anthropic’s position. Revenue of $17 billion in a single quarter, growing at more than 1,100 per cent year on year, is the kind of growth that justifies an extraordinary valuation. But commitments of $895.4 billion, including up to $120.5 billion payable to Elon Musk’s SpaceX for computing capacity through 2029, are the kind of obligations that make investors ask hard questions about whether the growth can be sustained, and at what cost. In its amended filing, Anthropic also added new risk-factor language warning that AI systems may soon accelerate their own development faster than safety checks can follow — a disclosure that this site reported earlier in the week, and one that sits awkwardly alongside the company’s founding promise to build AI safely.

The company held a developer day this week at which it unveiled a new Claude feature called “dreaming”, according to Al Jazeera, and it has continued to ship new products even as it prepares for the listing. But the AP deal is the more revealing announcement of the week, because it concerns the business rather than the technology. It shows a company that is learning, in public and under pressure, how to operate as a mature commercial enterprise rather than a research lab — and learning that maturity has a price, measured in licensing fees as well as in computing bills.

Anthropic was founded in 2021 by former OpenAI researchers led by its chief executive, Dario Amodei, and has raised billions from Amazon, which has invested $8 billion, and Google, which has invested more than $2 billion. Those backers are now also, in different ways, its competitors: Google builds its own Gemini models and runs its own licensing programme, while Amazon is Anthropic’s largest cloud partner and a major customer. The AP deal, modest as it is, sits at the centre of that web of relationships. It is a reminder that in the AI industry, the companies that compete most fiercely are also the companies that most depend on each other — and on the publishers whose work feeds them all.

Analysis: Why It Matters

The AP deal is small. Its terms are secret, its duration is unknown, and it may cover nothing more than a few weeks of election data. It matters anyway, because it is the first time the most principled holdout in the AI industry has paid for news — and because the reasons it finally paid reveal how the balance of power between technology companies and publishers is shifting. Anthropic did not change its mind because a court ordered it to, or because its lawyers told it that fair use would fail. It changed its mind because its product stopped working properly without the thing it had refused to buy. That is a more durable lesson than any legal ruling, and it is one that every other publisher will now try to apply.

1. The embargo worked — and publishers will use it again

The most important fact about this deal is not that Anthropic paid, but why it paid. Publishers blocked its crawlers, withheld their content, and waited for the moment when the absence of that content became a visible weakness. The tactic, which Adweek’s Mark Stenberg called a digital embargo, worked precisely because it targeted the one thing Anthropic could not solve with engineering: freshness. A chatbot can be trained on the open web, but it cannot be trained on tomorrow’s news, and on election night that gap was impossible to hide. The success of the embargo will not go unnoticed. Every publisher that has watched this negotiation will now ask what its own content is worth to an AI company at the moment of maximum need, and will price its access accordingly. The AP was the first to be paid; it will not be the last to be asked. The danger for publishers is that the tactic only works when they act together, and the history of the news industry suggests that collective action is hard to sustain once the first cheques start arriving. The AP’s deal may therefore be both a victory for publishers and the beginning of the end of their unity.

2. Fair use is a legal argument, not a business model

Anthropic’s long-standing position — that its use of publisher content was permitted under fair use — was never obviously wrong as a matter of law. A federal judge had, after all, ruled that training on lawfully acquired books was fair use. But the AP deal exposes the limits of fair use as a commercial strategy. Fair use may protect the act of training a model on text, but it does not oblige a publisher to let an AI company’s crawler into its website, it does not guarantee access to live data, and it does not give a chatbot the right to present a publisher’s election results as its own. As AI products have shifted from trained models to live services, the relevant question has stopped being whether training is lawful and started being whether access is available — and access is a matter of contract, not of copyright law. Anthropic’s experience suggests that the companies that will thrive in this new world are not the ones with the strongest legal arguments, but the ones with the best commercial relationships. OpenAI understood that earlier, and paid for it. Anthropic has now learned it the expensive way, on the most visible night of the year.

3. The AP deal is a template — and a warning

There is a reason the AP was the first publisher to license content to Google, the first to license to OpenAI, and now the first to license to Anthropic. Its business is licensing, its product is attribution, and its politics are, by design, nobody’s. That makes it the ideal partner for an AI company taking its first step into paid content — and it also makes it a misleading template. Most publishers are not wire services. They have audiences to protect, advertising to sell, and archives whose value depends on exclusivity. A deal that works for the AP may not work for a newspaper whose journalism is its only product, and a licensing fee that is trivial for a company with $17 billion in quarterly revenue may be derisory for a newsroom trying to pay its reporters. The AP deal should therefore be read as a proof of concept, not a settlement. It proves that Anthropic will pay, and that a deal can be struck quickly when both sides want one. It does not prove that the price will be fair, or that the model will scale to the hundreds of publishers whose work AI companies rely on every day. The negotiations that follow will be harder, longer and more contentious than this one — and the AP, having set the precedent, will be watching them closely.

4. The IPO is the real deadline

It is no coincidence that Anthropic’s first licensing deal arrived in the same fortnight as its IPO filing. A public company is judged by standards that a private company can ignore: its prospectus must disclose its risks, its investors must be persuaded that its business model is sustainable, and its every statement is tested against the record. A company that tells investors it may be worth $2 trillion while simultaneously telling publishers it does not need to pay for their work is making two claims that are hard to hold at once. The AP deal resolves that tension, at least symbolically: it gives Anthropic a concrete answer when investors ask how it will secure the data its products need, and it gives the company a story to tell about its maturity. But the IPO also creates a new pressure in the opposite direction. Once Anthropic is public, every licensing fee it pays will be visible in its accounts, and every publisher will know what its rivals were paid. The company’s willingness to pay will be tested not by one negotiation but by hundreds, conducted in the open, with the market watching. The AP deal was struck quietly, before the listing. The deals that follow will be struck in daylight.

5. The news business has found its leverage — for now

For two years, the news industry has watched AI companies build products on its work and wondered whether it had any leverage at all. The AP deal suggests that it does, and that the leverage is real-time information — the one thing AI companies cannot manufacture for themselves. A chatbot can summarise the past, but it cannot report the present, and the present is what readers actually pay for. That gives publishers something genuinely scarce to sell, and it gives them a reason to invest in the reporting that makes their content irreplaceable. But the leverage is also fragile. It depends on publishers maintaining their blocks, on readers continuing to value accuracy over speed, and on AI companies continuing to need live data rather than learning to generate it themselves. The history of technology suggests that today’s bottleneck is tomorrow’s solved problem. Publishers should take the AP deal as evidence that they can negotiate from strength — and as a warning that the strength may not last. The wise response is to use this moment to build licensing relationships that survive the technology changing, rather than to extract the highest possible price for the data that happens to be scarce today.

What to Watch Next

1. Whether the deal extends beyond the election: The most important unanswered question is whether the AP arrangement covers only the midterm period or becomes a standing licence. If Anthropic renews or expands it in the coming weeks, that will be the clearest signal yet that the company is building a licensing programme rather than patching a one-off gap.
2. Which publisher is next: With the AP precedent set, the focus shifts to the major newspapers and broadcasters that still block Anthropic’s crawlers. Watch for a deal with a large newspaper group, which would be a far bigger test of Anthropic’s willingness to pay — and of what it is willing to pay for.
3. The IPO roadshow: Anthropic is expected to begin marketing its Nasdaq listing in the coming weeks, with the offering set before Thanksgiving. Listen for how the company describes its content strategy to investors, and whether analysts press it on the cost of licensing at scale.
4. The first settlement payments: The first cheques from the $1.5 billion book settlement are due to authors and publishers on or before 15 November. How smoothly that distribution goes — and how loud the disputes over competing claims become — will shape how publishers think about dealing with Anthropic.
5. The copyright cases still in court: Anthropic still faces lawsuits from music publishers and news organisations, and the New York Times case against OpenAI and Microsoft continues. Any ruling in those cases could change the price of licensing overnight, in either direction.

Sources

• Adweek — Anthropic Strikes Its First Licensing Deal, With the AP, for Election Coverage (Mark Stenberg, 5 November 2026)
• Dealroom — Anthropic pays for news for the first time, licensing AP election data for Claude (5 November 2026)
• AI Weekly — Anthropic Licenses AP Election Data in First Publisher Deal (5 November 2026)
• TechCrunch — Authors Push Back as Publishers Claim Share of Anthropic’s $1.5 Billion Settlement (Anthony Ha, 6 September 2026)
• Open Class Actions — Anthropic $1.5B Copyright Settlement: Final Approval and Payment Update (Bartz v. Anthropic)