Anthropic has revised its Claude Usage Policy to prohibit sustained user abuse toward its models, ‘ban’ abusive conversations, and tackle misuse involving deceptive campaigns, election interference, weapons development and surveillance, while strengthening safeguards for high-risk applications and autonomous physical equipment.
The update introduces a prohibition on repeated, needless “abusive or cruel” behaviour toward its models. The rule targets extreme, repeated cruelty and excludes ordinary frustration, criticism, creative work, and model testing or research. It builds on an existing measure that allows Claude to “end rare conversations with persistently abusive users” on Claude.ai and Claude Code.
The revised policy will take effect on November 12. Anthropic said most changes spell out existing restrictions in response to the evolving capabilities of its models and emerging patterns of misuse.
The company has introduced a new section, “Do Not Engage in Deceptive Campaigns or Artificial Activity,” consolidating rules against deceptive campaigns, fake accounts and posts, and fabricated content.
Anthropic has also renamed its election policy section “Do Not Undermine Democratic Processes” and specified that Claude must not be used to deceive voters or disrupt elections such as spreading false information about candidates or voting procedures, impersonating candidates or election officials, and suppressing voter turnout.
The company has removed its blanket ban on personalized vote and campaign targeting while retaining restrictions on deceptive targeting and misuse of voters’ personal data.
The updated policy makes explicit that the existing prohibition on weapons development also covers software and components that enable weapons to function, including guidance and control systems. These changes come amid broader concerns highlighted in Anthropic’s September 2026 Threat Intelligence Report, which documented attempts to misuse its AI models.
Anthropic has also specified its restrictions on surveillance and criminal justice, prohibiting the use of Claude to track people without their consent, develop or improve surveillance tools, or recommend whom law enforcement should investigate, arrest or charge. The company said these changes make existing restrictions more explicit rather than alter its enforcement practices.
For high-risk applications affecting people’s health, legal rights, finances, livelihoods or access to essential services, Anthropic continues to require qualified human oversight and disclosure that AI was used.
Following the launch of its Model Hardware Standard, Anthropic has also introduced safeguards for models connected to autonomous physical equipment that could cause injury. These include requiring a qualified operator to monitor the equipment and be able to stop it.
Also Read: NYT Publisher: ‘Tech Giants Strip-Mine Original Journalism, Repackage Stolen Goods as Their Own’
