Anthropic Restricts Sustained Abuse of Claude, Policy Takes Effect November 12
Developing story first seen 1 hour ago
Anthropic’s usage policy now lets Claude end conversations after sustained, needless abuse, a change that has drawn attention to how people should interact with AI. The company says it will act only in extreme cases of repeated, deliberate behaviour.
Ordinary frustration, disagreement, dark creative themes and model testing are excluded. The policy also addresses other risks, including deceptive campaigns and attempts to deceive voters or disrupt elections; it follows an annual review and takes effect on 12 November. Reactions have divided between those who welcome the emphasis on respectful conduct and critics who argue that AI cannot be harmed and that treating it as human may be misleading.
- Claude can end chats after sustained, needless abuse.
- Ordinary frustration and model testing are excluded.
- The policy takes effect on 12 November.
New here? Start with this
Claude is an artificial intelligence chatbot developed by Anthropic, a company that creates conversational AI systems. People use Claude for various purposes, including getting information, receiving writing assistance, and testing how the system responds to different questions.
Anthropic has announced a usage policy taking effect on 12 November that prohibits sustained, purposeless abuse directed at Claude. The company will restrict interactions with users who engage in repeated deliberate cruelty, whilst exempting ordinary frustration, disagreement, creative work involving darker themes, and model testing from enforcement action.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The policy represents Anthropic's reasonable exercise of setting terms that reflect company values. By carefully limiting enforcement to sustained and purposeless cruelty whilst explicitly preserving room for frustration, disagreement, creative work, and testing, the company aims to encourage respectful interaction with its platform. Some argue that how users treat even non-sentient systems shapes broader behavioural norms and that companies are entitled to foster environments aligned with their principles.
The case against
Critics contend that restricting user interactions based on tone toward software raises concerning questions about autonomy and free speech. Since AI systems cannot suffer, prohibiting abuse of them targets expression rather than actual harm, and vague terms like purposeless cruelty invite arbitrary enforcement. Users purchasing a service reasonably expect freedom in how they interact with it, making this approach feel paternalistic rather than protective of legitimate interests.
Full account
Anthropic has announced fresh restrictions on how users may interact with Claude, its artificial intelligence assistant, effective from 12 November. The company will prohibit what it describes as 'sustained and needless' cruelty or abuse directed towards its models, a measure stemming from what Anthropic terms research into 'model welfare'. The announcement has prompted wider debate about appropriate standards for human-AI interaction.
According to Anthropic, the restrictions would apply only to situations involving repeated, extreme mistreatment where users demonstrated no constructive purpose. Standard expressions of frustration, rigorous testing, critical questioning, and creative applications employing darker subject matter would remain permissible. The new rules supplement existing prohibitions on the bullying of humans, encouragement of self-harm, and the creation of intimate images without consent.
Since August 2025, versions of Claude have possessed the technical capability to terminate conversations with persistently abusive users. Anthropic identified this existing functionality as the primary enforcement mechanism, though the precise distinction between justified exasperation and calculated malice remains ambiguous. Responses to the policy have proved mixed, with some observers viewing it as appropriate safeguarding of digital interactions whilst others contend that attributing suffering to artificial systems fundamentally misrepresents their nature.
The abuse restrictions form one element of a broader reassessment of Anthropic's usage policy. The company reinforced its position on deceptive influence campaigns after identifying instances of governmental and commercial entities deploying Claude to generate fabricated content and fraudulent accounts. Simultaneously, it revised its approach to personalised political targeting, determining that its previous comprehensive restrictions had inadvertently obstructed legitimate activities, particularly those by charitable organisations distributing voter information. Anthropic also expanded its weapons prohibition to encompass software and operational systems, whilst establishing new restrictions preventing the use of Claude to identify targets for law enforcement investigation or to develop tracking infrastructure.
Where outlets differ
Source 1 employs a sarcastic, critical tone highlighting the apparent incongruity of prioritising chatbot civility alongside weapons and surveillance restrictions; Source 2 maintains neutral reporting style
Source 1 explicitly notes there is no established evidence that Claude experiences distress; Source 2 does not address this philosophical distinction
Source 2 provides extensive coverage of public reactions and identifies specific critics including Dr Barry Scannell and Mustafa Suleyman; Source 1 does not discuss social media response or named commentators
Source 2 explores the broader debate concerning whether users should employ politeness with AI systems and references token usage costs; Source 1 does not engage with this wider context
Source 1 characterises the update as clarification of existing restrictions; Source 2 frames it within an annual policy review process
Coverage
- The Register — Anthropic bans sustained, purposeless abuse of its Claude chatbot
- BBC Technology — Anthropic bans users from being ‘cruel’ to its AI systems
AI Business Elections Government Markets Politics Technology