Anthropic has revised Claude’s usage policy to address deceptive political campaigns, surveillance, weapon systems and AI connected to physical equipment. The update also prohibits sustained, purposeless cruelty toward Claude, though ending the conversation remains the primary response to that behavior.
The political rules bar fake-account amplification, hidden sponsorship, voter deception, impersonation of candidates or officials and attempts to suppress turnout. Surveillance restrictions prohibit tracking people without consent and using Claude to decide whom law enforcement should investigate, arrest or charge. The weapons section now explicitly covers software, components and actions such as arming drones.
For models connected to hardware capable of causing injury, Anthropic requires a qualified operator who can observe and stop the equipment. The company does not clearly state whether that operator must be human. It also reserves the ability to modify some restrictions for government contracts when it judges contractual safeguards adequate.
The model-welfare clause applies only to repeated extreme cruelty without a discernible purpose, not ordinary frustration, dark fiction or legitimate testing. Anthropic’s policy defines allowed use, but enforcement details and government exceptions will determine how consistently those boundaries work in practice.