Anthropic: Data Buyer Profile, Program Terms, Payouts and Reputation

Anthropic, PBC is the Delaware public benefit corporation that develops the Claude models and acquires training data from the public web, from datasets it licenses or commissions, from consumer chats users allow for training, from synthetic data and, as a court record shows, from print books it bought and scanned. Its Form S-1 of 2026-10-26 names public web data, licensed or commissioned datasets and synthetic data as its sources.

Anthropic publishes no program that pays companies for data, and no payout or referral fee is located in the sources checked on 2026-11-09. DataPacts is an independent reference on AI training data licensing that takes no cut of any sale, and it covers Anthropic regardless of any referral program.

The figures and court statuses below were checked against Anthropic's filings, its published terms and court dockets; how DataPacts sources and dates every figure is set out on the methodology page.

How Anthropic Acquires Training Data: The Routes on Record

Anthropic's Form S-1 of 2026-10-26 names three training data sources: public web data, datasets licensed or commissioned from third parties, and synthetic data generated by its own models. The table adds the two routes that come from Anthropic's consumer terms and from a court record.

Table 1 compares the five routes on record, whether the source wording says the holder is paid, and the source (as of 2026-11-09).

"Paid to the holder" reflects the source wording only; DataPacts records no Anthropic price for any route.

Anthropic states that ClaudeBot collects web content that may contribute to model training and that its bots honor robots.txt rules. The bots, and how to allow, block or charge them, are covered under AI crawlers.

For licensed or commissioned datasets, no licensing agreement naming Anthropic as licensee is on record as of 2026-11-09, so no counterparty, scope or price can be stated.

Most routes in the table involve no payment to a holder. Only route 2 pays, and its terms are private, which leaves a company holding data without a public route to Anthropic.

Anthropic Company Identity

The registrant is Anthropic, PBC, a Delaware public benefit corporation with SEC CIK 0002130357, never "Anthropic Inc.". Table 2 lists the identity attributes as of 2026-11-09, each with its source.

The street address and headquarters city are not recorded in the sources checked.

Anthropic Program Terms, Payouts and Referral Fees, Dated

Anthropic publishes no program that buys or licenses data from companies and states no payout or referral fee; Table 3 lists every acquisition-related term it does publish, each with its check date.

The absence of a paid program and of a referral fee is what separates Anthropic from the buyers whose program terms compared show dated ranges, and the absence is a finding of the sources checked, not a statement by Anthropic.

Anthropic also restricts what its own customers may do with its output: the commercial terms bar using the services to train competing AI models. A holder who licensed data to Anthropic would therefore face a one-directional arrangement, with data flowing in and a restriction flowing out.

What a holder controls instead is limited to robots.txt rules naming ClaudeBot, the no-AI-training signal once it reaches the holder's content, and, for individuals, the consumer chat setting.

Print Books Anthropic Bought and Scanned: What the Court Record Shows

The Order on Fair Use of June 23, 2025 (N.D. Cal., Dkt. 231) records that Anthropic bought print books and scanned them into digital form for its library, and it held that digitizing those purchased print books was fair use. The court called the training use "exceedingly transformative".

Anthropic also looked for licenses. According to the same order, it hired a former head of partnerships for Google's book-scanning project in February 2024, and that hire emailed major publishers about licensing in spring 2024. The number of books bought, the spend and the scanning method are not in the court facts recorded here, and DataPacts does not estimate them.

The authors' class action over Anthropic's book copies, including the pirated copies the court separated from the purchased ones, ended in a USD 1.5 billion settlement with final approval on 2026-07-20 (Dkt. 680); claims, rulings and payments of the Bartz v. Anthropic settlement sit on the case page.

Claude Chats and Synthetic Data as Anthropic Training Sources

Since its consumer terms update of August 28, 2025, Anthropic uses chats from consumer Claude accounts for training when the user allows it, keeping that data for 5 years instead of 30 days. Anthropic's page gives the choice deadline for existing users as 2025-10-08, while launch press reported 2025-09-28; the two dates are shown here as a contradiction between sources.

Commercial services are excluded from that change: Claude for Work, the API (also through Amazon Bedrock and Google Cloud Vertex AI), Claude Gov and Claude for Education. Synthetic data generated by Anthropic's own models is a further route that pays no one, and Anthropic's S-1 lists it as a source.

The settings and their limits for users who want out are covered under opting out of AI training.

Anthropic Reputation: Lawsuits Over How It Acquired Data

Beyond the authors' book case, three suits challenge how Anthropic acquired training data: one by Reddit over automated access to its site and two by music publishers over song lyrics. All claims below are allegations, and each status is as of 2026-11-09.

Reddit, Inc. v. Anthropic PBC concerns automated access to Reddit content to train Claude. The federal court remanded it to San Francisco County Superior Court by order of 2026-03-30 (N.D. Cal. Dkt. 60), which found the state-law claims not preempted. Anthropic reports in its S-1/A that the state court narrowed the case in September 2026 to breach of contract, interference with contract and unfair competition.

Table 4 compares the three suits by court, docket, challenge and status as of 2026-11-09, with the case page for each.

In Concord I, the publishers allege that lyrics were used in training and appear in outputs, and a summary judgment ruling is pending. In Concord II, they allege that 20,517 works were used to train Claude, including acquisition by downloading, and on 2026-11-06 the court ordered the parties to meet and confer on Anthropic's motion to dismiss.

Is Anthropic a Realistic Licensee for Your Data? Benefits, Risks, Downsides and Alternatives

Anthropic offers no public route for a company to offer it data: it runs no data partnership program, states no payout and names no licensing counterparty on record, so a holder cannot apply to it the way it applies to a company data program.

An unsolicited email claiming to buy data on Anthropic's behalf deserves the checks that decide whether a data licensing email is legitimate before anyone replies. Anthropic's legal name is Anthropic, PBC, which gives a first check against the sender's details.

Benefits (as the record shows, none promised):

• The court record treats buying print copies as a lawful route.
• Anthropic sought publisher licenses in 2024.
• It is an SEC registrant with public filings.

Risks:

• Content on the public web may be crawled without payment unless blocked.
• Litigation over acquisition is pending.
• No license template or price is public.

Downsides:

• No published program, payout or referral fee.
• Counterparties of licensed or commissioned datasets are not disclosed.
• Anthropic's terms bar customers from training competing models on its services.

Alternatives:

• License through a company data partnership program or another frontier lab.
• License books or content directly through a publisher channel.
• Block or condition crawling.
• Do not license at all; see the alternatives to selling data, including not selling at all.

Anthropic Among Frontier AI Labs That Acquire Data

Anthropic is one of the frontier AI labs whose data acquisition DataPacts compares by route, from licensed catalogs to purchased books. The comparison of frontier AI labs as data buyers sets Anthropic beside the other labs.

• Every answer to who buys AI training data, including programs and intermediaries.
• OpenAI and Google as licensees with their own publisher and platform terms, and ClaudeBot as Anthropic's crawler.
• The five routes in how AI companies get training data, from crawling to synthetic data.

A holder weighing labs against vendors can use the guide to choosing a buyer for your data.

More About Anthropic

Human Feedback, Red-Teaming and RL Environments at Anthropic

In 2024 Anthropic ran an invite-only model safety bug bounty through HackerOne with rewards up to USD 15,000 for novel universal jailbreaks (announcement of 2024-08-08; current status not checked). It is one way labs acquire human data.

Its S-1 states that it suspended higher-risk RL environments after 2026 incidents in pre-release cybersecurity evaluations; the market for RL environments is covered on its own page.

Anthropic and EU Training Data Transparency

Anthropic is among the 22 signatories the European Commission lists for the General-Purpose AI Code of Practice (page last updated 2026-10-07, checked 2026-11-09). Anthropic's public training content summary under the EU AI Act Article 53 training data summary rules was not located in the sources checked.

Frequently Asked Questions About Anthropic

Does Anthropic buy training data?

Anthropic licenses or commissions datasets from third parties, according to its Form S-1, and it bought print books for scanning. It names no licensing counterparty in the sources checked.

Does Anthropic pay companies for their data?

No published Anthropic program pays companies for data, and no payout or referral fee is stated, checked 2026-11-09.

Does Anthropic train on API or business data?

Anthropic states that commercial services, including Claude for Work, the API, Claude Gov and Claude for Education, are excluded from its consumer training change. The API is also covered through Amazon Bedrock and Google Cloud Vertex AI.

Has Anthropic signed data licensing deals with publishers?

No licensing agreement naming Anthropic as licensee is on record as of 2026-11-09. The tracker of AI data licensing deals lists none.

Did Anthropic buy books to train Claude?

Yes: the June 2025 fair use order records purchased print books that Anthropic scanned. It held that digitizing them was fair use, and the wider topic is covered under books as AI training data.