August 13, 2026
Own it? Sure. Use it? Not so fast
If I own Claude's outputs why can't I train my own model on them?
Claude says you own the words, just don’t use them to build its rival
TLDR: Anthropic says you own Claude’s answers, but you can’t use them to train a rival chatbot unless it says yes. Commenters are calling foul, arguing that “ownership” sounds hollow if the company can still block what you do with the results.
Anthropic just dropped a very corporate version of “you can have it, but don’t get any ideas.” The company says users own what Claude writes for them, but there’s a giant catch: you’re not allowed to use those answers to build a competing AI chatbot without written permission. Anthropic says that’s about safety, control, and stopping people from using its own system to create a knockoff. In plain English: you can use Claude’s output for tools like sorting text, summaries, or search helpers, but not to make “Claude, but cheaper.”
And the comments? Absolutely not calm. The loudest reaction was basically: “If I own it, why are you telling me what I can’t do with it?” One commenter fired off the now-classic line, “If you can’t use the output to train, then you don’t own them,” turning the whole thing into a debate over whether “ownership” means anything if the rulebook is this long. Others went straight for the hypocrisy angle, accusing AI companies of training on the public internet and then slamming the door behind them. One person joked they added “No Rights for Hypocrites” to their site footer after seeing Claude scrape it anyway, while another called the whole thing “Robin Hood all over again”—except apparently everyone feels robbed.
The mood on the thread was a mix of anger, sarcasm, and meme energy. Anthropic says this is standard industry policy. The crowd’s response: standard doesn’t mean fair.
Key Points
- •Anthropic says users own Claude outputs, but its terms still restrict using those outputs to train or develop AI models without written permission.
- •The company states that training competing models on Claude outputs is prohibited under its terms, including supporting third parties doing so.
- •Anthropic says the restrictions are intended to preserve safety controls and prevent misuse of models trained on Claude outputs outside its oversight.
- •The article says permitted uses include training non-competing specialized tools such as sentiment analysis, categorization, summarization, information extraction, semantic search, and anomaly detection systems.
- •Prohibited uses listed in the article include general-purpose chatbots, open-ended text generation models, using outputs as training targets for models, and reverse engineering training methods.