Anthropic rewrites its Usage Policy: autonomous hardware needs a kill switch, personalized campaign targeting is back
The annual refresh, effective November 12, consolidates fake-account and influence-operation rules into one section, drops a blanket ban on personalized voter and campaign targeting that was blocking legitimate civic work, and requires a human-observable safe state for any Claude deployment connected to autonomous physical equipment.
Anthropic published its annual Usage Policy update on October 8, taking effect November 12. Anthropic says the refresh responds to new misuse patterns documented in its own September threat intelligence report -- influence operations, weapons development, and surveillance -- and that "most of the updates in the latest version are intended to clarify existing rules" rather than add new restrictions.
What actually changes for enterprise deployments
The most concrete new requirement is for any Claude deployment connected to hardware that takes autonomous physical actions: a qualified operator must be able to observe and stop the equipment, and the equipment must fail to a safe state if Claude is disconnected. The weapons section now explicitly covers software and components that make weapons work, plus arming drones or other autonomous vehicles -- Anthropic frames this as reflecting existing enforcement rather than a new line, which is a claim worth treating skeptically until it's tested against an actual case. The rewritten surveillance section prohibits tracking people without consent (real-time or via analysis of previously collected data) and bars Claude from deciding or recommending who to investigate, arrest, or charge, while explicitly carving out consent-based fraud monitoring, content moderation, journalism, and legal research as permitted.
The one rollback in an otherwise-tightening policy
Rules on fake accounts, fabricated news sites, and influence operations -- previously scattered across the elections, fraud, privacy, and disinformation sections -- are consolidated into a single "Do Not Engage in Deceptive Campaigns or Artificial Activity" section covering both political and commercial deception. But the renamed elections section ("Do Not Undermine Democratic Processes") drops the prior blanket ban on personalized vote and campaign targeting, which Anthropic says had been blocking legitimate civic work; targeting that relies on deception or misuse of personal data stays prohibited under the other sections. For a buyer assessing compliance risk rather than reading this as a philosophy statement, the practical items are the autonomous-hardware safe-state requirement (anyone running Claude against physical equipment needs to check their setup against it before November 12) and the narrower elections carve-out (anyone doing political-adjacent targeting work needs to re-read where the actual line now sits, since "personalized" is no longer automatically disqualifying).