Claude already has the ability to end abusive interactions on Claude.ai and Claude Code. Anthropic said ending a conversation will remain the main enforcement mechanism for this part of the policy.
The company also reorganized its election-related restrictions under a section called “Do Not Undermine Democratic Processes.” Claude cannot be used to distribute false information about candidates or voting, impersonate political candidates or election officials, or discourage people from voting.
At the same time, Anthropic removed its broad prohibition on personalized election and campaign targeting. The company said that rule had also blocked legitimate applications, including nonprofits creating voter materials in multiple languages and election officials notifying voters about ballot issues that need to be corrected.
Another section consolidates restrictions on deceptive influence campaigns. The policy prohibits using Claude to disguise who is behind a message, artificially amplify material through fake accounts, or create infrastructure for coordinated influence operations.
Anthropic said it has seen Claude used by state media organizations, government propaganda operations and commercial groups to operate networks involving fraudulent accounts and fabricated news websites.
The weapons policy has also been made more explicit. Existing restrictions now cover software and other components that are necessary for weapons to function, including work involving the addition of weapons to drones and other autonomous vehicles.
Anthropic similarly clarified its surveillance rules. Claude cannot be used to monitor people without their knowledge or consent, whether the surveillance happens in real time or relies on previously collected information. The model also cannot be used to determine or recommend which individuals should be investigated, arrested or prosecuted.
Those changes draw partly on abuse patterns Anthropic documented in a threat intelligence report covering attempted misuse between December 2025 and August 2026. The report described state-linked actors using Claude in activity related to biological weapons research, cyberattacks targeting Ukrainian government systems and surveillance of diaspora communities. Anthropic said it banned accounts connected to the identified activity and provided intelligence to authorities.
The revised policy also addresses Claude deployments involving hardware capable of taking physical actions independently. In those cases, Anthropic requires a qualified operator to be able to monitor the equipment and stop its operation when intervention is needed.
The policy changes broaden and clarify the boundaries Anthropic places around Claude as the company encounters more varied uses of its models, while preserving exceptions for ordinary criticism, research and legitimate civic activity.
This analysis is based on reporting from quartz & inc42.
Image courtesy of Anthropic.
This article was generated with AI assistance and reviewed for accuracy and quality.
