Anthropic Usage Policy Update: What Changes for Builders

Anthropic's usage policy update made headlines for banning "sustained and needless" cruelty toward Claude. That rule covers only extreme cases. For companies building on Claude, other changes matter more: new rules on deceptive campaigns, bots posing as humans, surveillance and machines that act in the physical world. The new policy takes effect on November 12, 2026.

Key Takeaways

• Anthropic announced its 2026 usage policy update on October 8. It takes effect on November 12, 2026.
• A new rule bans "sustained and needless abusive or cruel behavior" toward its models, but only in extreme cases. Normal frustration and testing aren't covered.
• The policy also adds rules against deceptive campaigns, including fake reviews, astroturfing and bots posing as humans.
• Weapons, surveillance and election rules were rewritten, and Claude-controlled hardware now needs an operator who can stop it.
• The policy covers end users who reach Claude through your app, so builders should update their own terms and checks before November 12.

What Changed in the Anthropic Usage Policy?

A lot more than the headline rule. Anthropic described it as its annual review, published as the 2026 Usage Policy update. Here are the main changes, as reported by MacRumors and Unite.AI:

Area	What changed
Treatment of models	New ban on sustained and needless abusive or cruel behavior toward Claude
Deceptive campaigns	New section against fake accounts, fabricated news sites, fake reviews, astroturfing and bots posing as humans
Elections	Renamed and narrowed: no deceiving voters or disrupting elections
Weapons	Now explicitly covers weapons software and components, and arming drones
Surveillance and policing	No tracking people without consent, no deciding who to investigate or arrest, no doxxing
Physical actions	A qualified operator must be able to observe and stop hardware Claude controls
Intimate imagery	New rule against creating, sharing or threatening to share non-consensual intimate imagery

Anthropic says most changes clarify existing rules rather than add new ones, MacRumors reported. The full text is in Anthropic's usage policy.

What Does the Rule on Cruelty to Claude Actually Say?

Less than the headlines suggest.

From Anthropic's update: The new rule prohibits "sustained and needless abusive or cruel behavior toward Anthropic's models." Anthropic says it's meant to "apply only in extreme cases," where users repeatedly act cruelly with no discernible purpose, Unite.AI reported.

It also says what's not covered: "common user frustration, pushback, dark creative themes, or model testing and research." So swearing at a chatbot that gets something wrong isn't the target. Neither is red-teaming, horror fiction or stress tests.

The penalty stays light. Claude's ability to end a conversation, available on Claude.ai and Claude Code, remains the main tool. Ending a chat blocks new messages in that one conversation. Other chats on the account keep working.

Why Did Anthropic Add It?

It grows out of research Anthropic calls model welfare.

In August 2025, Anthropic gave some Claude models the ability to end a small set of conversations, as a last resort, after attempts to redirect fail. It explained the move in a research post. It said it was unsure whether Claude has moral status, but wanted low-cost ways to reduce possible risks, according to MacRumors.

Anthropic also said Claude Opus 4 showed a "robust and consistent aversion to harm," and apparent distress when faced with abusive users. The feature isn't used when a person may be at imminent risk of harming themselves or others.

The new policy rule turns that rare practice into a written line in the terms.

When	What happened
August 2025	Some Claude models can end a small set of abusive chats
October 8, 2026	Anthropic announces its 2026 usage policy update
November 12, 2026	The updated policy takes effect

What Happens When Claude Ends a Conversation?

Not much, for most people. Here's how it works, based on Anthropic's 2025 research post and recent reports:

• Claude tries to steer the chat back first. It ends the chat only as a last resort, after several tries to move the talk somewhere useful have failed.
• A user can ask for it. Claude can also end a chat when the user asks it to.
• That one chat closes. No new messages.
• Everything else keeps working. Other chats are fine, and you can start a new one right away.
• You can go back. Earlier messages can still be edited and retried in a new branch.
• Safety comes first. Claude won't end a chat when someone may be at risk of harming themselves or others.

So the change is less a lockout and more a closed door on one conversation. Anthropic says ending the chat stays the main way it will enforce the new rule.

What Do Critics Say?

Plenty, and the debate is about more than manners.

• "You can't be cruel to numbers and maths," wrote Barry Scannell, a technology partner at Irish law firm William Fry, the BBC reported. "This level of anthropomorphisation of AI is harmful. It leads people to believe that it's something it's not."
• Mustafa Suleyman, Microsoft's AI chief, who criticized Anthropic in September for treating AI as if it were human, reacted with a melting face emoji, the BBC noted.
• Some critics worried the rule could blur the line between harm to software and harm to people and animals.

The other side: some users praised the change as encouraging good manners or taking "model welfare" seriously. And Anthropic stresses that the rule targets only extreme, purposeless abuse.

Our view: you don't need to settle whether an AI can be harmed to work with this rule. Very few people will ever hit it. The changes below, on the other hand, touch features that thousands of products already ship, from review widgets to outreach bots.

What Does the Anthropic Usage Policy Mean for Everyday Users?

Very little changes day to day. You can still vent when an answer is wrong. You can push back, argue and ask Claude to try again. Writers can still explore dark themes in fiction, and researchers can still test how the model behaves under pressure.

What's out is cruelty for its own sake, kept up again and again. Picture a user who spends a whole session abusing the model with no goal at all. That's the rare case.

If you're not sure where you stand, ask yourself one thing: is there a point to what you're doing? Testing, venting and fiction all have one, and the rule leaves them alone.

For most people, the other new rules matter more. Don't use Claude to write fake reviews. Don't use it to post as people who don't exist. And don't use it to track someone without their consent.

Which Anthropic Usage Policy Changes Matter Most for Builders?

The rules that touch common product features. Five deserve a close look.

• Bots posing as humans. The deceptive campaigns section bans bots that pretend to be people. If your agent sends messages, label it as AI where the other person would expect to know.
• Fake reviews and astroturfing. Tools that generate reviews, testimonials or grassroots-style posts need clear limits.
• Election content. Political tools must not deceive voters or disrupt elections. Note that the blanket ban on personalized campaign targeting was removed, but deceptive targeting and misuse of voters' data remain banned under other sections.
• Surveillance. Products that track people, build profiles or support policing decisions face sharper limits, though uses like fraud monitoring and journalism stay allowed.
• Physical actions. If Claude controls hardware, such as robots or drones, a qualified operator must be able to watch it and stop it.

Anthropic said it added the deceptive campaigns section after finding state media, government propaganda offices and commercial firms running networks of fake accounts and fabricated news sites with Claude, according to MacRumors.

Does the Anthropic Usage Policy Cover Your Users?

Yes, and that's the part builders most often miss.

The policy applies to "anyone who can submit inputs to Anthropic's products or services," including "through authorized resellers or passthrough access," Unite.AI reported. If customers type prompts into your product and those prompts reach Claude, their use must follow the policy too.

That makes you the first line of enforcement. In practice:

• Your terms of service should match the policy.
• Your monitoring should catch obvious misuse, such as mass fake-review generation, before it becomes a problem with your account.
• Your product design should make misuse harder, for example by limiting bulk posting features or requiring human review for public content.

Our look at AI agent permissions covers how to limit what agents can do on a user's behalf.

Anthropic Usage Policy Checklist Before November 12

About a month is enough if you start now. A practical checklist:

Check	Who owns it
List every feature that sends Claude output to third parties, such as emails, posts or reviews	Product
Add AI disclosure where messages could be mistaken for a human	Product and design
Update your terms of service and acceptable use rules	Legal
Review any political, surveillance or identity features against the new sections	Legal and product
Confirm an operator can stop any Claude-controlled hardware	Engineering
Tell staff what happens if Claude ends a chat in Claude.ai or Claude Code	Team leads

That last item is small but avoids confusion. Anthropic describes the conversation-ending feature for Claude.ai and Claude Code. It hasn't said the same applies to apps built on the API, so check the docs before you design around it.

Should You Be Polite to AI?

The policy reignited an old debate, and the honest answer is that it's mostly up to you.

Some argue that "please" and "thank you" waste tokens. Others find polite, clear prompts get better answers. When asked about the cost of users saying please and thank you, OpenAI's Sam Altman once said it was "tens of millions of dollars well spent," the BBC noted.

For teams, the practical rule is simpler. Write prompts that are clear, specific and calm. That helps the model, helps your colleagues read your prompts later, and stays far away from anything the new policy covers.

There's a simple test. Would you be happy if a colleague read the prompt? If yes, you're fine.

Plain prompts also tend to work well: say what you want, who it's for and what a good answer looks like. Skip the insults and the flattery. Neither adds much.

How Does the Anthropic Usage Policy Compare With Others?

Every major AI provider has a usage policy, and they share a lot: bans on weapons, abuse of people, fraud and election interference. Anthropic's rule on treatment of the model itself is unusual, which is why it drew so much attention.

For builders, the practical point is the same across providers. Read the policy of each model you use, map it to your features, and keep a record of the review. If you switch or add providers, repeat it. Our guide to AI governance covers how to keep that review going without slowing teams down.

How Van Data Team Helps Teams Stay Compliant

We help teams build AI products that follow their providers' rules without slowing down. That means mapping features to usage policies, adding disclosure and review steps, setting up monitoring for misuse and testing agents before launch.

The Anthropic usage policy update is a good reason to review your product now. If you want a second pair of eyes before November 12, our guide to AI agent evaluation is a good place to start.