You own Claude’s outputs. You still can’t train a competitor on them.

Anthropic gives you ownership of Claude’s outputs, but it bars you from using them to train or develop AI models without the company’s written permission. The rule is set out in an Anthropic help-center article on Claude dated March 16. The article calls such restrictions “standard practice across the AI industry.”
Next to the blanket ban sits a narrower test. You can use outputs to train models “that don’t compete with Anthropic’s own models.” The article’s examples are specialized classifiers and tools. What the terms forbid is training models that compete with Anthropic’s, and helping a third party do that is also a violation. Outputs can also go into your own applications: they can power product features, generate content for your customers, analyze and structure your data, and improve internal workflows.
Anthropic gives two kinds of reasons. The first is safety. Models trained on Claude’s outputs “won’t have our safety measures,” the article says, which could lead to harmful systems. Anthropic adds that it has no visibility into where such models are deployed: “we cannot monitor how these distilled models are used or prevent misuse.” Distillation here means training a model on another model’s answers so it picks up that model’s behavior at a fraction of the cost.
The second reason is commercial, and the article says so directly. Customers who generate outputs to train competing models are “essentially using our infrastructure and investment to build direct competitors to our service,” Anthropic writes, adding: “Like other software and service providers, we expect that our services won’t be used to undermine our product offerings.” Of the two, this is the one Anthropic can actually enforce through a contract. A safety layer that disappears after distillation is a risk. A competitor built on your API spend is a breach of the terms.
Anthropic has also framed distillation in larger terms. Dario Amodei has argued that industrial-scale distillation could narrow China’s gap with the US frontier to a matter of months, and he called for targeted measures against the operations that do it.
This isn’t the first time Anthropic has drawn a line around a rival. In 2025 it cut Windsurf’s first-party access to Claude 3.5 Sonnet and Claude 3.7 Sonnet while OpenAI was reported to be acquiring the coding tool for $3 billion. “With less than five days of notice, nearly all of our first-party capacity to all Claude 3.x models has been cut off,” Windsurf CEO Varun Mohan said, per DevX. Anthropic co-founder Jared Kaplan explained that decision in one line: “I think it would be odd for us to be selling Claude to OpenAI.”
Up next
OAuth login to MCP servers is fixed
Release 2.1.231 lists exactly one fix: OAuth login to MCP servers was failing with a redirect URI mismatch error. It hit servers that use a pre-registered OAuth client, Slack included.
Related stories
- Claude Code keeps your session data locally for 30 days
- Anthropic paid $740,000 to a firm led by Rubio’s ex-aide
- Linear and Ramp users can pay for Claude without an API key
- WSJ: Anthropic on Trump’s list of AI national champions
- Hong Kong and mainland funds barred from Anthropic IPO
- Claude expands MCP Apps support with app tools and more
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

