The Horizon
From November 12, being cruel to Claude breaks Anthropic's rules
This one sounds like a joke, but it’s real: Claude can hang up on you, and now it’s in the rules. From November 12, 2026, Anthropic’s usage policy says users must not engage “in sustained and needless abusive or cruel behavior toward our models.”
What happened
Anthropic announced the change on October 8 as part of its 2026 Usage Policy Update. It covers Claude users, including on Claude.ai and Claude Code, according to Notebookcheck and PCWorld.
What is and isn’t covered
The rule is aimed at extreme cases: users who repeatedly act cruelly toward the model with no purpose. Per the reporting on Anthropic’s announcement, these are not covered:
- ordinary frustration and pushback (like telling Claude it’s wrong)
- dark themes in creative writing
- testing and research on the model
PCWorld points out that the exact line between frustration and cruelty isn’t spelled out.
What actually happens
The main enforcement is something Claude can already do: end the conversation. Some Claude models got that ability in August 2025, as part of Anthropic’s work on what it calls model welfare. Anthropic says Claude’s ability to end these interactions “will remain the primary enforcement mechanism.” When a chat ends you can’t send more messages in it, but your other chats are not affected, and you can start a new one.
The policy’s general clause still lets Anthropic warn you, or limit, suspend or end your access, for a violation, as with any other rule in the policy.
Why it matters
Most AI policies are about protecting people from the model. This one is partly about how people treat the model. Anthropic doesn’t claim Claude has feelings. Its model welfare research treats it as an open question whether AI systems could have anything like experiences that matter, and it has chosen to be careful while that question is open.
For most of us nothing changes. Shout at the bug, tell the model it’s wrong, keep working. Just don’t spend an evening being cruel to it for fun.
Sources
The Horizon
Engineering leadership for the AI era, every other week.
One honest piece every two weeks on AI in software delivery, platforms and leading teams. No spam, unsubscribe anytime.