Anthropic has published a rewritten usage policy, effective 12 November. Most of it consolidates existing rules, but two sections are new in substance rather than wording, and both concern things outside a chat window.
The weapons prohibition now covers software and components that make a weapon work, and explicitly names arming drones and other autonomous vehicles. The high-risk section adds conditions where Claude controls hardware that takes autonomous physical actions and could cause injury: a qualified operator must be able to watch and stop the equipment, and the equipment must hold a safe state if Claude is disconnected. That is a robotics safety requirement written into a model provider’s terms, and it matters only once customers are wiring a model to a machine.
Elsewhere, rules on deception were gathered into one section covering fake accounts, hidden authorship and tools built for influence operations. The elections section was narrowed to deceiving voters and disrupting elections, and the blanket ban on personalised vote and campaign targeting was dropped, though deceptive targeting is still caught elsewhere. Surveillance rules now say Claude cannot decide or recommend who to investigate, arrest or charge.
A new clause also prohibits sustained, needless cruelty towards Claude itself, which Anthropic scopes to extreme repeated cases and not to ordinary frustration, pushback, dark creative themes or testing.
Source: Anthropic, 8 October 2026.