Anthropic is drawing a line on how people treat its AI. Starting November 12, 2026, the company's revised usage policy will explicitly forbid users from directing sustained and needless cruelty or abuse toward Claude.
The rule applies everywhere Claude runs, including web chats at Claude.ai, the official mobile apps, and the free tier that recently gained access to Claude Haiku 5.5. Anthropic announced the changes on October 8, framing the cruelty clause as an edge-case guardrail rather than a speech filter. Venting routine frustration, pushing back against wrong answers, exploring dark creative themes, and conducting safety research remain fully allowed.
Enforcement leans heavily on an automated mechanism already running behind the scenes. In August 2025, Anthropic gave select Claude models the authority to shut down a session if someone persists with abuse after multiple warnings. When that happens, the model locks the specific conversation so no further messages go through. It does not lock you out entirely: other saved threads stay open, and you can spin up a fresh chat immediately. Anthropic noted that this automated exit remains its main tool, though the company keeps the right to issue warnings, throttle queries, or ban accounts for severe violations.
Stricter Drone and Police Tracking Rules
The updated document also tightens several real-world safety boundaries. On the surveillance front, users cannot use Claude to track individuals without consent, whether through live feeds or by running queries against historical location logs. Anthropic also barred the tool from making recommendations on who should be targeted for police investigations, arrests, or criminal charges. Carve-outs still protect consensual fraud detection, moderation workflows, legal research, and journalism.
On the battlefield, Anthropic expanded its existing weapons prohibition. The updated text specifically names the software and individual parts that run weapon platforms, with a direct ban on systems designed to arm drones.
Political rules saw a slight adjustment in the opposite direction. Anthropic scrapped an older blanket prohibition on personalized political campaign targeting, provided campaigns do not deceive voters or mishandle private voter files. To counter automated influence operations, the policy gathered its rules against fake account rings and fabricated news sites into a single section.
For everyday work in Claude Code or routine writing tasks, your day-to-day workflow will look identical. The updated restrictions take effect across all Claude tiers on November 12.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.