Anthropic bans cruel behavior toward Claude in policy update

Artificial intelligence startup Anthropic has updated its usage policy for the first time in over a year. The revision introduces rules targeting high-risk misuses such as election interference, surveillance, and weapons development, while also formally barring sustained and needless abusive or cruel behavior toward its Claude model.

Anthropic stated that terminating conversations remains the primary enforcement mechanism for abusive behavior, expanding on an August policy that permitted Claude to end chats with persistently harmful users. The company noted that the rule applies only to extreme cases where users repeatedly act cruelly without a discernible purpose, and does not apply to testing, dark creative themes, or ordinary user frustration.

The updated rules consolidate restrictions against deceptive political or commercial campaigns, explicitly forbidding attempts to disguise message sources or inflate content using fake accounts. The policy also bars voter deception, impersonation of candidates or election officials, and actions aimed at suppressing voter turnout.

Anthropic broadened its prohibition on weapons to encompass software and components that control weapons or arm autonomous vehicles, following observed attempts to use Claude for guidance systems. In addition, the company banned unauthorized tracking of individuals, building surveillance tools, and using Claude to decide or recommend investigations, arrests, or charges in law enforcement, though terms may be adjusted in specific government contracts with adequate safeguards.

A new clause mandates that whenever Anthropic models connect to hardware capable of causing physical injury, a qualified operator must be able to monitor and halt the equipment. The welfare provisions align with ongoing internal research into model experience, though industry peers have pushed back, including Microsoft AI stating in September that it rejects legal personhood or rights for artificial intelligence.