anthropic tightens the rules of use by Claude by explicitly prohibiting long-term abusive or cruel behavior towards his artificial intelligence. Applicable from November 12, 2026, this update also strengthens restrictions on electoral disinformation, weapons and surveillance.
Claude can shorten abusive conversations
The new rule targets users who attack Claude for no apparent reason. Anthropo specifies that it only refers to extreme situations in which a person repeatedly demonstrates cruelty. Frustration, criticism, conflictive exchanges, dark fictional scenarios and model testing are still permitted.
This decision extends an experiment launched in August 2025: Claude can now end a conversation with an interlocutor who persists in his abusive behavior. Anthropic confirms that this disruption will remain its primary means of enforcing the rule. The update does not specify a new automatic penalty for this type of behavior, although the general policy provides for measures up to the closure of an account.
In any case, the question of the well-being of AI models remains controversial. Anthropic explores the possibility that AIs may have some form of subjective experience, without claiming that Claude possesses consciousness. This approach generates disagreement in the artificial intelligence industry, where many officials believe that current systems do not have feelings or rights comparable to those of humans.
Disinformation, weapons and surveillance in the viewer
Anthropic takes advantage of this revision to group together various prohibitions and specify prohibited uses. These are the main changes:
- Deceptive campaigns: prohibition of creating fake reviews, artificially manipulating opinions, creating fake sites or using bots that impersonate humans.
- Elections: prohibition of misleading voters or disrupting the conduct of the vote, in particular by disseminating false information about the vote.
- Arms: explicit extension of restrictions on the software and components necessary for the operation of weapons, as well as the assembly of drones and unmanned systems.
- Law enforcement: Claude cannot recommend who to investigate, arrest, charge or prosecute. Doxxing and the development of surveillance tools are also prohibited.
- Physical actions: When equipment operated by Claude presents a risk of injury, a qualified person must be able to monitor its operation and stop it.
- Intimate images: prohibition of creating, distributing or threatening to distribute intimate images without consent, as well as developing the tools intended to produce them.
The updated policy also specifies the conditions applicable to sensitive uses in the fields of health and finance, as well as systems capable of physically acting autonomously. Most of these changes clarify existing restrictions or consolidate previously scattered rules. Anthropic indicates that it has observed attempts to use Claude to feed networks of fake accounts and media, which notably motivated the strengthening of rules against deceptive influence campaigns.
