A Claude user says Anthropic flagged the user’s account for “training a competitor” and that “anything touching RFT, SFT, or training larger models” now gets the account blocked, in an X post that drew 204,262 views and a 330-point Hacker News thread.
Separately, three Anthropic documents restrict using its services or outputs to train competing models, by our count: its Commercial Terms, its Consumer Terms and its Usage Policy. As of 1 November we found no public Anthropic statement on the account.
Short explainer · Anthropic terms and a user’s training block
A vertical news Short summarising this story. The user’s account is their own; the terms text is from anthropic.com/legal.
Business Pill · Open Weights Explained
A short explainer on renting a model versus running open weights yourself. It teaches the general idea only and says nothing about this user or Anthropic.
The key insight: As we read the documents, the line Anthropic draws is about outputs and competing models; the user’s complaint is that, by their account, the training data came from an open-source model and Claude only submitted jobs and ran analysis.
In the post the user says they “never distilled Claude”, that “rollouts came from a strong open-source model” and that “Claude just helped submit jobs and run analysis”.
The user says the result was being blocked, “Opus 5.5 downgraded to Opus 5”, or “the session killed outright”. These are the user’s claims; we found no public Anthropic response to them.
The Hacker News thread linking the post had 330 points and 38 top-level comments when we read it on 1 November. No comment in it identifies itself as coming from Anthropic.
Two user-reported restrictions on Claude in two days, set against Anthropic’s published terms. User claims are the users’ own (X post by @jiahaoshao1, 31 October; Hacker News; FourWeekMBA’s 30 October report); the terms text is from anthropic.com/legal. The count of three documents is ours.
What Anthropic’s Terms Say
The Commercial Terms say in section D.4 that a customer may not “access the Services to build a competing product or service, including to train competing AI models” except as expressly approved by Anthropic.
The Consumer Terms list as a prohibited use: “To develop any products or services that compete with our Services, including to develop or train any artificial intelligence or machine learning algorithms or models”.
The Usage Policy bars users from using outputs “to train an AI model (e.g., โmodel scrapingโ or โmodel distillationโ) without prior authorization from Anthropic”.
Where This Fits
It is the second user-reported restriction on Claude in two days. On 30 October we reported that Claude Code users described a new rule against recreating games, and Hacker News also carried Tom’s Hardware’s report on Claude blocking game cloning and reverse-engineering.
The Business Lever: Terms as a Moat
As we read the documents, the terms protect Anthropic’s models from being used to train rivals. The user’s complaint is about the other side of that line: research automation where, by the user’s account, the training data came from an open-source model and Claude only ran the jobs.
The Structural Read
As we read the terms, the Commercial Terms and Consumer Terms restrict using the services to build or train competing models, and the Usage Policy separately restricts training on Claude’s outputs without authorization.
As we read the post, the user describes three different responses: being blocked, a downgrade from Opus 5.5 to Opus 5, and sessions killed outright; the post does not say which happened when.
As we read the timing, this is the second user-reported restriction on Claude we have covered in two days, after the 30 October report on a rule against recreating games.
Anthropic Usage Policy, prohibited uses
“Use outputs to train an AI model (e.g., โmodel scrapingโ or โmodel distillationโ) without prior authorization from Anthropic”
Three Implications
FOR TEAMS AUTOMATING RESEARCH WITH CLAUDE The user says work touching RFT, SFT or training larger models was blocked even though, by their account, Claude did not supply the training data.
FOR READING THE RULES Three Anthropic documents restrict training competing models or training on outputs, by our count; none of the quoted passages says how restricted use is detected.
FOR READING THE CLAIM The account is one user’s public post; we found no public Anthropic statement on it as of 1 November.
The framework puts it plainly: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Anthropic’s terms keep its model outputs closed to use in training rivals. The framework is the owner’s argument and does not discuss this user or this post.
We could not verify the user’s account history, what triggered the flag, or whether the blocks and model downgrade the user describes were automated or manual. The terms texts do not say how Anthropic detects restricted use. We found no public Anthropic statement on the post as of 1 November.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
A Claude user says Anthropic now blocks anything touching model training on their account; Anthropic’s Commercial Terms, Consumer Terms and Usage Policy all restrict using its services or outputs to train competing models, and we found no public statement from the company.
A note on sourcing. The account of blocks and downgrades is the user’s own, from a public X post; we could not verify it. The terms quotations are from Anthropic’s published documents as read on 1 November. The count of three documents is ours. We did not contact Anthropic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
