Anthropic published a new version of the rules for using its Claude AI on Thursday, Oct. 8. The rules, which Anthropic calls its Usage Policy, set out what people and companies may and may not use Claude for. Anthropic says most of the changes clarify existing rules, adding examples now that Claude takes on longer, more independent work, and that they draw on misuse it saw over the past year in propaganda campaigns, weapons development and surveillance. The new version takes effect on Nov. 12.
On weapons, the rules now say plainly that the ban covers the software and parts that make weapons work, as well as arming drones and other self-driving vehicles; Anthropic says people have recently tried to use Claude to build guidance and control software for weapons. On surveillance, Claude may not be used to track people without their consent, whether live or by sifting data already collected, to build surveillance tools, or to decide or recommend whom to investigate, arrest or charge in a police or court case. Tracking people agree to, such as fraud monitoring, stays allowed, along with content moderation, journalism and legal research. Anthropic says neither change alters what it already enforces.
The rules on fake accounts and fake news sites are gathered into one new section that covers political and commercial campaigns alike, including any effort to hide who is really behind a message, and the elections section now focuses on deceiving voters or disrupting elections, such as spreading false information about how to vote or impersonating election officials. Anthropic also dropped a blanket ban on tailoring messages to individual voters, which it says blocked useful work such as nonprofits writing voter information in other languages; targeting that relies on deception or misused personal data stays banned under other sections.
Two rules are new. When Claude is connected to machines that act on their own and could hurt someone, someone qualified must be able to watch the equipment and stop it, and the machine must be able to hold a safe state, such as stopping safely, if Claude disconnects. Anthropic also now bars "sustained and needless abusive or cruel behavior" toward its AI, which it says applies only in extreme cases where users are repeatedly cruel for no clear purpose, not to everyday frustration, pushback, dark creative writing or testing. Anthropic has said it is "highly uncertain" about Claude's potential moral status, meaning whether its wellbeing could matter at all, and since August 2025 it has let Claude end rare conversations with persistently abusive users, who can start a new chat right away; that will remain the main way the rule is enforced.
Anthropic has said what the cruelty rule does not cover but not what does, and it is not yet clear whether breaking it could cost anyone their account. The company also spelled out how it enforces its list of places where Claude is not offered: the ban covers people physically there, companies based there, and, as it has since last year, companies mostly owned or controlled from there.