Anthropic Updates Usage Policy to Ban Model Abuse and Election Interference

New Prohibitions Target Harmful Interactions and Misinformation

Anthropic announced on Thursday that it has revised its usage policy, adding explicit bans on election meddling, the development of weapons software, surveillance applications, and—most notably—a prohibition against sustained verbal abuse directed at its AI models. The company clarified that the restriction on abusive dialogue is intended for extreme, repetitive cruelty without any legitimate purpose.

Since an August update, Claude has been programmed to terminate exchanges that become persistently harmful. The revised policy now goes further, preventing users from continuing such conversations once the model has signaled it will end them.

The policy notes that ordinary user frustration, creative dark humor, or research testing do not trigger the new rule, emphasizing that only repeated, purposeful cruelty is targeted.

Background of the Policy Shift

The timing of the change follows Anthropic’s outreach to prominent religious scholars, during which the company has sometimes suggested that Claude might possess consciousness or a soul.

Measures Against Deception and Electoral Disruption

Additional sections of the updated policy forbid users from launching deceptive campaigns, such as operating fake accounts or fabricating news outlets with Claude’s assistance. A dedicated segment titled “Do Not Undermine Democratic Processes” outlines rules that bar efforts to mislead voters or interfere with elections.