Friday, 9 October 2026
Anime, manga, and games, with a take · A Yukimedia publication
← all stories other 4 sources · 1h ago ·

Anthropic Bans Persistent Abuse Of Claude In Usage Policy Update

A usage policy that assigns the model itself protection from user conduct moves AI rules beyond output restrictions toward the question of the model's own treatment, a question Anthropic says it is researching but has not resolved.

Key Facts

  • Anthropic revised its Claude usage policy on October 8, 2026, with the new rules taking effect on November 12, 2026.
  • The revision prohibits sustained and unnecessary abusive or cruel behavior directed at Claude.
  • Anthropic states the prohibition does not cover ordinary dissatisfaction or pushback, dark creative works, or model testing and research.
  • The update consolidates rules on deceptive campaigns, elections, fraud, and disinformation into one section and adds drone arming to prohibited weapons uses.
  • The policy requires human oversight for health, finance, and physical-device uses of Claude, with a way to stop dangerous actions.

Reporting from 4 sources: Denfaminicogamer, ASCII.jp, GameBusiness.jp, GIGAZINE.

Anthropic Bans Persistent Abuse Of Claude In Usage Policy Update

Anthropic revised its usage policy for the Claude AI model on October 8, adding a prohibition on sustained and unnecessary abusive or cruel behavior directed at the model. The revised policy takes effect on November 12, 2026. The prohibition covers extreme cases in which a user repeatedly aims cruel language or conduct at Claude without clear purpose. Anthropic states that ordinary expressions of dissatisfaction or pushback, dark creative works, and model testing or research fall outside the scope. The update also consolidates existing rules on deceptive campaigns, election interference, fraud, and disinformation into one section, adds the arming of drones and other autonomous vehicles to the prohibited uses for weapons development, and clarifies restrictions on surveillance systems. For health and finance uses, and for cases in which Claude operates physical devices, the policy requires human oversight and a mechanism to stop dangerous actions, plus the ability for devices to stay safe if the connection to Claude is lost.

The revision also codifies a mechanism already running on Claude.ai and Claude Code, where the model can end a conversation that turns persistently abusive, according to ASCII.jp. Anthropic says ending the conversation remains its primary response.

GameBusiness.jp ties the clause to Anthropic's "Model Welfare" research, which examines whether AI may have experiences equivalent to consciousness or suffering. The company introduced a feature in August 2025 letting Claude end conversations with users who repeatedly use malicious language. Anthropic states it is uncertain whether AI has consciousness or suffering.

Other sections changed beyond the abuse clause.

  • "Do Not Engage in Deceptive Campaigns or Artificial Activity" consolidates existing rules on elections, fraud, privacy, and disinformation. Building tools or infrastructure for influence operations also falls under the ban.
  • The elections section is renamed "Do Not Undermine Democratic Processes." It bars disinformation about candidates or voting methods, impersonation of candidates or election officials, and use aimed at suppressing turnout.
  • The blanket prohibition on voter and campaign targeting for individuals is removed, a change GameBusiness.jp links to nonprofits that give voters information in multiple languages. Deceptive targeting and misuse of voter data stay banned.
  • Devices tied to Claude must stay safe if the connection drops.

GIGAZINE notes the update is described as "mostly for the purpose of clarifying existing rules," and that Anthropic revises the policy as needed in response to model evolution and user feedback.

Synthesized by Yomimono from the 4 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.

Claude Anthropic

More on this

Sources