Anthropic has written a ban on sustained cruelty toward its Claude chatbot into its usage policy, a rule that took effect on November 12

1

3

. It lands just as a study found a third of American chatbot users have sent at least one insult

2

1

.

Source: Tom's Guide

Source: Tom's Guide

What the ban covers

The new clause prohibits "sustained and needless abusive or cruel behavior" toward Claude

1

. Anthropic says the target is extreme, repeated cruelty without clear purpose, not everyday frustration, disagreement, creative writing or legitimate testing of the system

1

3

.

The company grounds the rule in precautionary research on AI welfare rather than any evidence that Claude suffers

1

3

. It acknowledges that whether AI deserves moral consideration at all is an open debate

1

3

.

What happens to a user who crosses the line

Claude has been able to end conversations with persistently abusive users since August 2025, as part of the same welfare research, and ending the chat remains the primary response

1

3

. A user whose conversation ends can generally start a new one, give feedback or edit and retry their messages; there is no automatic lockout

1

3

.

Beyond that, Anthropic's broader usage policy allows warnings, throttling, suspension or termination of access

1

3

. The policy covers the website, Claude Code, the API, cloud services, resellers and apps from other providers; none of these options has been reported as actually imposed

1

3

.

The study behind the timing

Searchable, an AI visibility platform, studied at least 8 million opt-in American chatbot conversations with ChatGPT, Gemini and Anthropic's Claude over the year to September 2026

1

. It ran from September 2025 to September 2026

1

. The single vendor study found that 33% of US chatbot users sent at least one message containing a swear word, an insult or a slur

2

1

3

.

The commonest triggers were the chatbot failing: ignored instructions accounted for 19% of abusive messages, wrong answers for 18% and misunderstood requests for 15%

2

1

.

And venting appears to backfire: across the study, civil complaints, including those aimed at Claude, were resolved more often than abusive ones

2

1

3

. The exact figures differ between outlets, one putting end-of-conversation resolution at 41% for polite corrections against 31% for abusive ones

2

and another reporting the split the other way round

1

, so the safe reading is the direction, not the decimals. "Telling a chatbot exactly what went wrong gives it something to fix, while an insult gives it nothing to work with," said Chris Donnelly, Searchable's CEO and co-founder

1

.

No causation was established, and it gives no separate abuse rate for Anthropic's chatbot

1

3

.

What is still unsettled

Critics worry the rule pushes people to treat AI as human and weakens human control

3

. Where the line sits between criticism and abuse, and how legitimate testing of the system is treated, remain unsettled

3

. How Anthropic enforces the rule through the API, what penalties it has handed down and how commercial users can appeal are questions no one has answered yet

3

.

© 2026 TheOutpost.AI All rights reserved