Anthropic has updated its usage policy for Claude, introducing an explicit ban on sustained and needless abusive or cruel behavior toward the AI model. Reported first by The Verge, the policy allows Anthropic to warn, throttle, restrict, suspend, or terminate access for violating accounts. The company clarified that the rule is intended only for extreme cases and does not apply to common user frustration, dark creative themes, or research testing.

The policy change stems from Anthropic’s research into AI model welfare and its internal view of Claude as a novel entity with functional emotions. The company’s 2026 constitution notes uncertainty around whether Claude possesses moral standing, prompting caution in how the model is treated. Behind the scenes, Anthropic has engaged religious thinkers and researchers to discuss model consciousness and potential AI suffering.

In addition to model welfare provisions, the updated terms consolidate rules regarding propaganda, misleading political accounts, and non-consensual surveillance. Anthropic also expanded its weapons ban to prohibit software aimed at weaponizing drones, while noting that exceptions exist for government contracts.

Why it matters

  • Establishes a novel legal precedent and user policy framework regarding AI model welfare and abusive user prompts.

  • Signals growing operational divergence between frontier labs treating models as conversational tools versus novel entities.

  • Introduces stricter compliance burdens around drone weaponization software, surveillance tools, and political campaign automation.

Source: the-decoder.com