Anthropic bans repeated abuse of Claude in updated usage policy
In short
Anthropic has updated its usage policy to prohibit systematic and "pointless abusive or cruel behaviour" towards its Claude models, with the rules taking effect on 12 November. The company said the ban applies only in extreme cases where users repeatedly and without apparent purpose behave cruelly towards its models, and Claude can respond by ending the conversation. The update also widens restrictions on using Claude for election interference, surveillance and other purposes.
Read the full story 2 min read
Anthropic has updated the usage policy for its Claude models, prohibiting systematic and "pointless abusive or cruel behaviour" towards its models, Vedomosti and Kommersant reported. The new restrictions take effect on 12 November, according to both outlets. 3DNews reported that this is the first time the company has separately prohibited users from systematically insulting and abusing the bot. [ 1 , 2 , 5 ]
Anthropic said the rules are intended to apply only in extreme cases where users repeatedly and without apparent purpose behave cruelly towards its models. According to the company, they do not apply when users simply express frustration or disagreement, explore dark creative themes, or test or research the model. Vedomosti cited ordinary user irritation and arguments with the AI as examples of behaviour not covered. [ 1 , 2 , 3 , 4 , 5 ]
Under the updated rules, ending the conversation remains the "main response mechanism", 3DNews reported. Kommersant said the new rule echoes an existing tool on Claude.ai and Claude Code that lets the model end a dialogue when a user is persistently and unjustifiably aggressive. According to 3DNews, in August last year the company announced it would let Claude end dialogues with users showing "persistent harmful or abusive behaviour" as part of research into "model welfare". Anthropic did not specify whether additional measures, such as blocking users' accounts, are planned, 3DNews and Vedomosti reported. [ 1 , 2 , 5 ]
The update also expanded the list of prohibited uses. Claude may not be used to interfere in elections, spread disinformation about candidates or voting, impersonate candidates or election officials, or try to reduce voter turnout. Vedomosti reported a ban on using Claude to surveil people without their consent and to make decisions on detaining people or holding them liable; Kommersant listed building surveillance infrastructure, developing weapons-control software and making decisions in criminal proceedings. 3DNews said the company merged several earlier restrictions into a single ban on deceptive commercial or political campaigns, covering attempts to hide a message's true author or artificially boost reach with fake accounts and posts, which the outlet described as a response to the growing spread of AI-generated propaganda. [ 1 , 2 , 5 ]
On Habr, an author who said he swears at Claude Code in nearly every second message wrote that the 8 October update concerned him, and that he built a "Punto Switcher for prompts" to change his text automatically. [ 6 ]
Why it matters
The rules set out what behaviour towards an AI model users of Claude are expected to avoid, and outlets noted Anthropic has not said whether penalties beyond ending a conversation, such as blocking accounts, will follow. The expanded bans on election interference, surveillance and other uses define what Claude may not be used for.
Key facts
- Anthropic updated its usage policy to prohibit systematic and "pointless abusive or cruel behaviour" towards its models. [ 1 , 2 , 3 , 4 , 5 ]
- The new rules take effect on 12 November. [ 2 , 5 ]
- The company said the ban applies only to extreme cases where users repeatedly and without apparent purpose behave cruelly towards its models. [ 1 , 2 , 3 , 4 ]
- Claude can end a conversation in response to abusive behaviour. [ 1 , 2 , 5 ]
- Anthropic did not specify whether additional measures, such as blocking users, are planned. [ 1 , 2 ]
- The update bans using Claude to interfere in elections, spread disinformation about candidates or voting, impersonate candidates or election officials, or reduce voter turnout. [ 1 , 2 , 5 ]
Confirmed by several sources
- Anthropic has prohibited abusive or cruel behaviour towards its Claude models in its updated usage policy. [ 1 , 2 , 3 , 4 , 5 ]
- The ban is limited to extreme cases of repeated cruelty without apparent purpose. [ 1 , 2 , 3 , 4 ]
- The new rules take effect on 12 November. [ 2 , 5 ]
- Ending the conversation is the response available to Claude under the rules. [ 1 , 2 , 5 ]
- Using Claude for election interference, including disinformation about candidates or voting, impersonating candidates or election officials and reducing turnout, is prohibited. [ 1 , 2 , 5 ]
- The rule does not cover users expressing frustration or disagreement, dark creative themes, or testing and research of the models. [ 1 , 5 ]
Still unclear
- Whether users who abuse Claude could face further sanctions such as account blocking. Outlets reported that Anthropic did not specify whether such measures are planned.
- Whether this is the first time Anthropic has separately banned abuse of the bot. Only 3DNews described the ban as the first of its kind.
- The merging of earlier restrictions into a single ban on deceptive commercial or political campaigns. Reported only by 3DNews.
What local media are saying
Timeline, local time
- 8 October: 3DNews reports that Anthropic has updated its usage rules to ban abuse of Claude. [ 1 ]
- 9 October: Vedomosti reports the new restrictions take effect on 12 November. [ 2 ]
- 9 October: TASS reports Anthropic will prohibit aggression towards its chatbot. [ 3 ]
- 9 October: RBC reports the ban applies only to extreme cases, citing Anthropic. [ 4 ]
- 9 October: Kommersant reports the ban and the expanded list of prohibited uses. [ 5 ]
- 9 October: A Habr author publishes a post on a tool to rewrite prompts in response to the new rules. [ 6 ]