Anthropic bans 'abusive or cruel behavior' towards Claude

What it claims

Anthropic has updated Claude's usage policy for the first time in over a year. For anyone building on Claude, the operative details are the scope limits and exact wording below.

Abuse toward Claude. The ban covers "sustained and needless abusive or cruel behavior." Anthropic says it applies "only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose." It explicitly does not apply to:

• ordinary user frustration or pushback
• dark creative themes
• model testing and research

Red-teaming, adversarial evaluation and fiction workloads are therefore carved out by name.

Enforcement. Claude ending the conversation, introduced last August, remains the "primary enforcement mechanism." Anthropic declined to say whether account bans or other measures could follow.

Surveillance. The wording is broader than "tracking dissidents":

• Tracking people without consent is banned "whether it happens in real time or through analysis of previously collected data." Retrospective analysis of existing datasets is covered.
• Claude cannot "decide or recommend who to investigate, arrest, or charge." This is a ban on recommendation, not only on final decisions.
• Using Claude to "build or improve tools designed for surveillance" is prohibited. This reaches development work, not just runtime use.

Weapons. The ban now extends to "software and components that make weapons work," which covers guidance and control software, not only weapons themselves.

Influence campaigns. Several previously "scattered" restrictions are consolidated into one rule. It prohibits "efforts to obscure who is behind a message or amplify content through fake accounts or posts." Election rules cover misinformation about candidates or voting procedures, impersonating candidates or election officials, and turnout suppression.

Physical hardware. When Claude is "connected to hardware that takes autonomous physical actions and which might be capable of causing injury," a "qualified operator must be able to observe the equipment and stop it if needed."

Limits

• Operator requirement. The policy says "qualified operator," and the article notes it is not clear whether that operator must be human. Robotics integrators should not assume an automated watchdog either satisfies or fails the rule until Anthropic clarifies.
• Government exemption. Exemptions for "certain governmental customers" depend on whether, "in Anthropic's judgment," contractual restrictions and safeguards are adequate. No criteria are published.
• Health and finance. These are named as areas of change, but the article gives no detail on them. Check the policy text directly.
• Source. This is a news report quoting the policy, not the policy itself. Verify exact clauses against Anthropic's published usage policy before relying on them for compliance.