Anthropic updated its usage policy with clearer restrictions on election manipulation, weapons, surveillance, autonomous hardware, and repeated cruelty toward Claude.
Anthropic has revised its usage policy to address several risks it says have become more prominent, including election interference, weapons development, surveillance, and sensitive health and financial uses. The update also bars repeated cruelty toward Claude when it has no discernible purpose. Anthropic says this restriction is intended for extreme cases, not ordinary frustration, disagreement, dark creative work, or model testing and research.
For that conduct, Anthropic identifies ending conversations as its primary enforcement measure. The company did not tell The Verge whether it might also impose measures such as user bans. The policy change follows an earlier Anthropic announcement that Claude could end conversations with persistently harmful or abusive users as part of the company’s model-welfare research.
The revision consolidates restrictions on deceptive commercial and political campaigns, including attempts to hide who is behind messages or amplify material through fake accounts. Its election rules also cover misleading voters, impersonating candidates or election officials, and efforts to suppress turnout.
Anthropic says weapons-related restrictions now explicitly cover software and components that enable weapons, as well as arming drones and other autonomous vehicles. The company has also clarified its surveillance prohibitions: these include tracking people without consent, using Claude to recommend who should be investigated or charged, and building or improving surveillance tools. Anthropic said some government contracts could receive modified rules if it judges that adequate contractual limits and safeguards mitigate the risks.
For AI connected to hardware capable of taking physical actions that could cause injury, the policy requires a qualified operator to be able to observe the equipment and stop it if necessary. The update therefore combines rules for model interactions with limits on how Claude may be used in political activity, monitoring, weapons-related work, and physical systems.