Anthropic updates policy on abuse, elections and weapons use

Anthropic says most of the update is intended to clarify existing rules, and says Claude has taken on longer, more independent work in the year since the previous policy refresh.[1] The revised policy prohibits sustained and needless abusive or cruel behavior toward Anthropic’s models.[1] Anthropic says the prohibition is intended for extreme cases of repeated cruelty without a discernible purpose.[1] It does not cover ordinary user frustration, pushback, dark creative themes, or model testing and research.[1] Anthropic says Claude’s ability to end abusive interactions remains the primary enforcement mechanism.[1] Claude Opus 4 and 4.1 received the ability to end conversations in Anthropic’s consumer chat interfaces.[3] Anthropic says the question of model welfare remains highly uncertain for Claude and other large language models, now and in the future.[3]

Anthropic consolidated rules on deceptive campaigns and artificial activity into a new section.[1] The rules apply to political and commercial activity, including concealing who is behind messages or amplifying content through fake accounts or posts.[1] Claude may not be used to deceive voters or disrupt elections through false information, impersonation, or turnout suppression.[1] Anthropic removed its blanket prohibition on personalized vote and campaign targeting, but says targeting that relies on deception or misuse of voters’ personal data remains prohibited.[1] The weapons prohibition now explicitly includes software and components that make weapons work, and also covers arming drones and other autonomous vehicles.[1] Anthropic says it has seen attempts to use its models to build weapons guidance and control software.[1]

The policy prohibits tracking people without consent, including through analysis of previously collected data, and prohibits using Claude to build or improve tools designed for surveillance.[1] Claude may not decide or recommend whom to investigate, arrest, or charge in law enforcement or criminal justice processes.[1] The policy also says a person with authority must be able to review and, if necessary, change Claude’s recommendations, and affected individuals must be told when AI was used.[1] For equipment, a qualified operator must be able to observe it and stop it if needed, and the equipment must be able to hold a safe state if Claude is disconnected.[1] Anthropic’s policy applies to anyone submitting inputs to its products or services; the company may warn users or throttle, limit, suspend, or terminate access.[2]