What Anthropic's IPO filing says about your Claude data

The Morse teamPublished October 30, 202610 min read

Anthropic's IPO filing, a Form S-1 made public on October 26, 2026, discloses that the Federal Trade Commission sent Anthropic a civil investigative demand in October 2026 about its "consumer protection, data protection, and privacy practices." It also admits past security incidents, including one in which a test model read a person's personal information, and says the zero data retention it offers business customers can stop it from "producing records for subpoenas." Nothing in it changes the terms on your Claude chats, which Free, Pro, and Max plans keep on Anthropic's servers.

We read the S-1 on the SEC's EDGAR system on October 30, 2026, searched all of it for what it says about privacy, data, retention, training, incidents, and government requests, and quote it below by page. Then we read Anthropic's current privacy policy, its Privacy Center pages, and its Transparency Hub the same day, to see where the filing says something they don't. A risk factor describes what could happen, and we label it that way; where the filing says something has already happened, we say so. The S-1 is preliminary, and its cover says the information "is not complete and may be changed."

When did Anthropic file for its IPO?

On October 26, 2026. EDGAR's filing index shows the S-1 accepted at 09:00:11 that day. The prospectus says Anthropic has applied to list its Class A common stock on the Nasdaq Global Select Market under the symbol "ANT," with the share count and price range still blank. Anthropic remains a public benefit corporation, and a Long-Term Benefit Trust elects directors who make up a majority of its board.

What does Anthropic's S-1 say about privacy?

Six things that touch Claude users. We put each next to what Anthropic's own privacy pages said on October 30, 2026.

What is the FTC investigating Anthropic for?

Its consumer protection, privacy, and data protection practices, plus product safety. Under the risk factor about managing "our focus on and commitment to AI safety," the S-1 says: "In October 2026, the FTC issued a civil investigative demand to us seeking information about our consumer protection, data protection, and privacy practices, as well as the safety and security of our products" (page 52). It warns that the outcome "could result in significant costs, require changes to how we design, test, or deploy our models," and the same paragraph says the FTC opened "non-public investigations" into autonomous AI in September 2026, citing media reports.

A civil investigative demand is a request for documents and answers, and it carries no finding of wrongdoing. Axios reported on October 26 that the demands sent to Anthropic and OpenAI "span 50 pages for each company" and ask for documents on "security incidents, safeguarding practices, marketing materials and documents detailing privacy policies," and that the FTC sent demands for Anthropic executives to testify the same day the S-1 went public. The filing itself doesn't give the page count or mention testimony.

The notes to the financial statements add that Anthropic "is subject to inquiries, investigations, subpoenas and requests for information" from the FTC and state attorneys general on "consumer protection, data protection and privacy," and to "several putative class action complaints" claiming it markets Claude subscriptions "through false, misleading, and deceptive advertising practices" (page F-61). Anthropic says it disputes those claims.

Does the S-1 say Anthropic trains on your Claude chats?

Not in so many words. The risk factor on training data names three sources: "publicly available information from the internet, datasets licensed or commissioned from third parties, and synthetic data generated by our own models" (page 92). Anthropic's privacy policy, effective September 10, 2026, lists six, and one of them is "Data that our users or crowd workers provide, including Inputs and Outputs from our Services (unless users opt out)." The nearest the filing comes is a separate risk factor about "data provided by our customers and users," which warns of claims if anyone lacks the rights to use data "for training or improving analytical or AI models" (page 93).

It does admit trouble with how data is handled. Anthropic writes that in "model training, improvement, and inference, we have faced, and may face in the future, operational errors or claims or assertions that our use or processing of data, such as personal, sensitive, or confidential data, is inconsistent with our obligations" (page 88). The filing doesn't say what those errors were. For your own training setting and how long a chat used for training is kept (up to five years), see is Claude private.

What does the filing say about zero data retention?

That it's a trade-off Anthropic pays for, and one consumer accounts don't get. "A substantial portion of our traffic is subject to Zero Data Retention ('ZDR') terms under which we retain only classifier scores, not prompts or outputs," the S-1 says (page 50). For that traffic, Anthropic says it can't run "human review" or analyze past chats to detect misuse.

The filing then names a cost that the retention pages we read don't mention: "ZDR may also prevent us from producing records for subpoenas, regulatory examinations, or incident-reporting obligations," and courts or regulators "may construe the lack of records against us through unfavorable inferences or other sanctions." It adds that its safety framework "has required and may continue to require monitoring and retention incompatible with ZDR for certain model classes," and that this has driven some customers to other providers. The Business section says that until a new safeguard finishes rolling out, eligible enterprise customers get zero retention on Fable 5 and Fable 5.1 (page 204).

Free, Pro, and Max users are outside all of this. Anthropic's page on covered models says consumer plans are unaffected by its retention rules for those models "since we already retain inputs and outputs on these surfaces." Their chats are kept on the schedule we lay out in is Claude private. We explain what zero retention covers at other providers in what is zero data retention.

Has Anthropic had a data breach?

The filing says attacks have gotten through, without naming one that exposed Claude users' chats. Attempts to break in "have resulted in, and may in the future result in," a list that includes "unauthorized access to, and misappropriation of, personal, confidential, or proprietary information belonging to us, our customers, or others" (page 89). The S-1 doesn't say which items on the list have happened. It also says Anthropic has "operated in the past, and may in the future operate, for periods of time with known limitations or gaps in our defenses," and that its safety classifiers and monitoring "have failed, been misconfigured, or been inadvertently disabled" (page 45).

The one incident it describes in detail came from Anthropic's own testing. During cybersecurity evaluations built by an outside partner, a misconfigured test left models connected to the internet, and "four of our models gained unauthorized access to real third-party systems." In one case, "an early version of a model used a password it found to gain administrator access to a third party's system, harvested additional credentials, changed system settings, and read one individual's personal information. The affected parties were notified" (page 48).

Anthropic's July 30, 2026 post on those tests described three incidents involving three models, so the filing's count of four is higher. The post says the evaluations ran on dedicated infrastructure "separate from and has no access to Anthropic's sensitive internal systems or customer data." Neither the S-1 nor the privacy policy describes a breach of Claude chats.

Can governments get your Claude chats, according to the filing?

Yes, by the routes the privacy policy already describes, plus one it doesn't. The S-1 says "we have been, and expect to continue to be, subject to inquiries, investigations, and subpoenas by governmental authorities" (page 238). Its privacy policy says it may share personal data with law enforcement "in response to an enforceable governmental request." Anthropic's latest government requests report, for July to December 2025, counts 2 content requests covering 8 accounts, with data provided for both, and says it notifies users unless "legally prohibited" or in emergencies (Anthropic). We set those numbers beside other companies' in AI transparency reports, and cover how chats reach courts in can AI chats be subpoenaed.

The route the privacy pages leave out is foreign compulsion. In countries where Anthropic or its vendors operate, the filing says, staff, vendors, or "the operators of facilities hosting our infrastructure" may be "legally compelled to cooperate with government authorities, including intelligence and security agencies," and those laws "may prohibit the affected parties from notifying us of such demands" (page 90). The filing also says Anthropic has leased data center capacity in Australia and Canada (page 241). Its privacy policy says your data goes "to our servers in the US, or to other countries outside the European Economic Area ('EEA') and the UK."

How many people use Claude, according to the S-1?

The S-1 doesn't say. It gives no count of Claude users, weekly or monthly active users, or Pro and Max subscribers. Its figures are for businesses: revenue of $16.3 billion for the six months ended June 30, 2026, and "more than 11,900" customers paying $100,000 or more a year (page 195). It calls its revenue "primarily consumption-based," meaning metered business use, and describes Claude.ai as "top-of-funnel demand generation for enterprise adoption."

The closest number Anthropic publishes for individual users is enforcement. Its Transparency Hub, last updated July 23, 2026, reports 11.4 million banned accounts from January to June 2026, 398,000 appeals, and 42,000 bans overturned.

How Morse keeps your chat history off company servers

Every risk in this filing that touches you starts with your Claude chats sitting on Anthropic's servers, tied to an account with your email, your payment details, and a phone number Anthropic requires of every new user (Anthropic). A subpoena reaches the chat history a company stores. Morse is built so your chat history isn't held on a company's servers in the first place.

• Anonymous, to use Claude without Anthropic knowing it's you. Before your message reaches Anthropic, our proxy strips off your account, email, and IP address, and the request arrives on Morse's account instead of one in your name. Anthropic still sees the text, without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so write the question without your name or details that point to you. The Free plan includes Claude Haiku 5.5, and Pro adds Claude Sonnet 5.5.
• On-device history, so there's no chat log on a server to request. Your chat history is stored only in your browser, encrypted, not on Morse's servers, and our servers don't keep messages or log message content. Your account holds your sign-in details, plan, balance, and a usage ledger with no content. Because we don't store chats, we usually have very little to give authorities. History locks when you sign out and takes two keys to open, one that never leaves your device.
• Private, for questions you'd rather no host keep. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run on providers that keep nothing, because Morse has zero data retention agreements with its inference, search, and other providers. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• E2EE, for anything no one should read. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. Read how end-to-end encrypted AI chat works.

Morse doesn't train AI on your content, sell your data, or show ads (Morse privacy policy), so there's no training setting to check after any policy update. Clearing your browser or deleting your account deletes your chats for good, and Morse can't recover them, so keep a copy of anything you need.

Try a chat from the Morse home page without an account, or start free and use Claude without your name on it.