Election targeting rules change
Anthropic has announced a revised Usage Policy for Claude that removes its blanket prohibition on personalized vote and campaign targeting. The company says that restriction also covered legitimate civic activities, including preparing voter information in other languages and sending notices about correcting ballot problems. Publication and implementation are separate: Anthropic says the updated policy takes effect November 12.
The removal comes with continuing restrictions. Anthropic says targeting involving deception or misuse of voters’ personal data remains prohibited. It has also brought previously scattered rules together into a section covering deceptive campaigns, whether political or commercial. According to the company, that section covers concealing who is behind messaging, using fake accounts or posts to amplify content, and building infrastructure for influence operations.
Human oversight and physical safeguards
For uses affecting health, legal rights, finances, livelihoods or essential services, Anthropic says its existing oversight requirements remain in place. A qualified person must have authority to review Claude’s recommendations and change them when necessary, and affected individuals must be informed that AI was used. The company says it has rewritten this section to make clearer which recommendations fall within those requirements.
The update adds requirements for models connected to equipment that acts autonomously and could cause injury, according to Anthropic. A qualified operator must be able to watch the equipment and halt it when necessary. The equipment must also maintain a safe condition if its connection to Claude is lost. These requirements concern the physical operation of connected hardware.
A narrowly scoped rule on model abuse
Anthropic is also adding a prohibition on sustained, unnecessary abusive or cruel conduct toward its models. The company describes the intended scope as extreme cases involving repeated cruelty without an identifiable purpose. It says ordinary frustration, disagreement, dark creative material, and model testing or research fall outside that prohibition, limiting how broadly the new rule is meant to apply.
The company connects that addition to an existing capability: Claude models can end rare conversations involving persistent abuse on Claude.ai and Claude Code. Anthropic says ending those interactions will remain the primary enforcement mechanism. The announcement therefore introduces a policy prohibition alongside a response that the company says its models could already take.
Sources & context
- 2026 Usage Policy updateAnthropic
Single-source report; claims attributed to the source, not independently verified. Source feed date (published): 2026-10-08T00:00:00.000Z. Page publication date: 2026-10-08T17:00:00.000Z. Study date: not established. Consulted: 2026-10-08T19:00:17.162Z.
AI assisted with preparing and checking this article. The responsible editor accepted this version for publication following an assisted review of facts, sources and attribution.
Related reading
