Anthropic's new Claude rules ban surveillance tools

The cruelty rule got the laughs. The surveillance rule is the one that can end a product.

Anthropic published a revised Usage Policy on Thursday, effective November 12. Anthropic says most changes clarify existing rules, adding examples for Claude's longer, more independent work and for misuse in influence operations, weapons development and surveillance.

Seven changes, two that bite

Anthropic's summary lists these changes:

• Tracking people without consent is now prohibited, whether in real time or by analyzing data already collected, and building or improving surveillance tools is banned.
• In law enforcement and criminal justice, Anthropic writes that "Claude cannot be used to decide or recommend who to investigate, arrest, or charge."
• The weapons ban now explicitly covers software and components that make weapons work, and arming drones or other autonomous vehicles.
• The elections section is renamed "Do Not Undermine Democratic Processes" and narrowed to deceiving voters or disrupting elections. The blanket ban on personalized vote and campaign targeting is removed.
• A new section, "Do Not Engage in Deceptive Campaigns or Artificial Activity," covers political and commercial deception alike, including fake accounts and tools for influence campaigns.
• When Claude is connected to hardware that acts on its own and could injure someone, a qualified operator must be able to observe and stop it, and it must hold a safe state if Claude disconnects.
• The ban on unsupported regions now names entities majority-owned or controlled by people or companies in those regions.

Then the one everyone shared: a ban on "sustained and needless abusive or cruel behavior toward our models."

Anthropic says it covers only extreme cases of repeated cruelty with no discernible purpose. Frustration, pushback, dark creative themes and model testing are out of scope. The main enforcement is Claude ending the conversation on Claude.ai and Claude Code.

The policing line is wider than it looks

Yes, a rule about being rude to a chatbot is an easy joke. It's also the rule least likely to touch you.

The surveillance language reaches further. "Decide or recommend" covers a lead-ranking tool for investigators as much as a tracking pipeline.

Anthropic says its enforcement practice hasn't changed and the rewrite makes the rule explicit. Fair enough. But an explicit rule is the one that gets quoted back to you when an account is reviewed.

The elections change runs the other way. Personalized vote and campaign targeting was banned outright. Now it's out of the elections section, while deceptive targeting and misuse of voters' personal data stay prohibited under other sections, per Anthropic.

Check your product against it

If you build on Claude, ask three questions:

• Does your app follow a person's location or activity without consent, from live data or old records?
• Does it rank or flag people for investigators, prosecutors or police?
• Does Claude drive equipment that moves?

A yes to any of them means reading the full Usage Policy, linked from Anthropic's post, before November 12. The post itself is a summary of the changes.

This is the second rules change from Anthropic this week, after it split Claude's cyber access into three tiers.

Watch November 12

That's the deadline. After it, watch for the first public enforcement case under the surveillance language. That's where the line actually gets drawn.

Read the policy. Then reread your roadmap.