Anthropic released an updated usage policy on Oct. 8, adding a new rule that bans the sustained and unnecessary abuse or cruel treatment of its models. The revised policy will take effect on Nov. 12 and also folds together language covering fake-account influence operations, elections, weapons, and surveillance.
Rule is limited to extreme cases, with ending the chat still the main response
The new provision appears in the policy section that bars cruel, abusive, or psychologically harmful conduct. Anthropic said it applies only to extreme situations in which a user repeatedly treats the model cruelly and there is no apparent reason for doing so. It does not cover ordinary user frustration, arguing with the model, dark fictional or creative material, or testing and research.
Anthropic said the rule matches how it has already been handling the issue. Since August 2025, Claude has been allowed to end a small number of conversations with users who keep abusing it, a measure currently in place on Claude.ai and Claude Code. That type of abuse is the main target of this update, and ending the conversation will remain the primary enforcement method.
At the time, Anthropic said it was not sure whether Claude holds moral status, but wanted to reduce possible risks to the model in a low-cost way.
New deceptive activity section covers fake accounts and false amplification
Anthropic said most of the changes are clarifications of existing rules. The company pointed to a threat intelligence report released last month, which disclosed that state media, government propaganda units, and commercial firms had used Claude to operate fake-account networks and fake news websites.
Those prohibitions had previously been spread across sections on elections, fraud, privacy, and misinformation. The updated policy combines them into a chapter that bans deceptive activity or the creation of false engagement. The restriction applies to both political and commercial uses and covers hiding the actor behind a message, amplifying content through fake accounts or fake posts, and building tools used to carry out influence operations.
Election section renamed, blanket ban on personalized voter and campaign targeting removed
The elections section has been renamed to focus on conduct that undermines democratic processes. Anthropic said the policy is aimed at blocking the use of Claude to deceive voters or interfere with elections, including spreading false information about candidates or voting methods, impersonating candidates or election officials, and suppressing turnout.
At the same time, Anthropic removed an earlier blanket prohibition on personalized voter and campaign targeting. The company said the old language also blocked legitimate uses, such as nonprofit groups using Claude to draft voter information in other languages or election officials sending ballot cure notices.
Weapons, surveillance, and law-enforcement limits were also rewritten
In the weapons section, Anthropic explicitly added software and components that enable weapons to function, along with attaching weapons to drones and other autonomous vehicles. The company said it had recently seen attempts to use the model to develop weapon guidance and control software, and that the revised wording reflects how it had already been enforcing the rules.
The surveillance and law-enforcement section was also rewritten. Tracking someone without consent is prohibited in all cases, whether in real time or through analysis of existing data. In law-enforcement or criminal justice settings, Claude cannot be used to decide or recommend who should be investigated, arrested, or prosecuted, and it cannot be used to build or improve surveillance tools.
Tracking with the subject’s consent, such as fraud monitoring, as well as content moderation, news reporting, and legal research, remains allowed.
Hardware safeguards and supported-region restrictions were updated too
The revised policy also says that when Claude is connected to hardware capable of autonomous physical action that could cause harm, a qualified operator must be able to observe the equipment and stop it if needed. The equipment must also remain in a safe state if Claude loses connection.
Anthropic separately updated its supported regions page. Under the new language, people located in unsupported regions, organizations registered or headquartered in those regions, and entities majority-owned or effectively controlled by individuals or organizations from those regions are not allowed to use Claude.