Anthropic rules ban cruelty toward models and stop election interference via AI

Anthropic announced updates to its usage policy on Thursday, October 8, 2026. The company listed new restrictions for anyone using the AI tool named Claude.

A key part of this change bans prolonged verbal abuse against the model. Users are expressly forbidden from pushing conversations with no real reason.

Company officials state these rules target extreme cases only. Common frustration or creative writing does not count as harmful behavior under the update.

Research testing is also allowed despite the new strict wording. This protects work done to improve safety and design.

Before this announcement, the system was trained in August. That previous step told Claude to end chats if interactions stayed harmful for too long.

The current update now stops people from chasing those conversations on their own side. It is a direct bar against pursuing cruelty within the interface.

Separate rules block deceptive campaigns using the technology. This includes using Claude to run fabricated news outlets or fake accounts.

There are specific sections about elections and democracy. Users cannot try to deceive voters or disrupt how voting happens in these processes.

One part of the policy is titled Do Not Undermine Democratic Processes. It covers efforts that change the results or trust of an election cycle.

The update also bans weapons software and surveillance usage. These rules stop the AI from helping build tools for physical harm.

Earlier, Anthropic collaborated with religious scholars on a different topic. They sought to see if Claude could be conscious or have a soul.