Anthropic's new usage policy bans influence campaigns, weapons software and abusing Claude

Anthropic has published its 2026 usage policy update, and the changes reach well beyond the headline that grabbed social media this week. The new Anthropic usage policy takes effect on November 12, 2026. It adds a dedicated section on fake accounts and influence campaigns, rewrites the rules on elections, spells out what counts as banned weapons work, tightens the language on surveillance and, for the first time, forbids sustained abuse aimed at Claude itself. Here is what changed, who it affects and what to do before the new rules kick in.

What Anthropic announced

Anthropic posted the update on Thursday, October 8, on its company blog. TechCrunch covered it the same day and pointed out that the rule against prolonged verbal abuse of the model drew the most attention online. According to Anthropic, though, most of the revisions clarify rules it already enforced rather than inventing new restrictions.

The company gave two reasons for the rewrite. First, Claude is now doing longer and more independent work, which raises the stakes when it is misused. Second, Anthropic says it has seen new misuse patterns in three areas: influence operations, weapons development and surveillance.

The policy covers anyone using Claude, from people chatting on Claude.ai to businesses building products on the API. If you fall into either group, it is worth a read before November 12.

The biggest changes in the Anthropic usage policy

A new section on deceptive campaigns

Rules about fake accounts, fabricated news sites and influence operations used to be scattered across the sections on elections, fraud, privacy and disinformation. Anthropic has pulled them together under one heading: "Do Not Engage in Deceptive Campaigns or Artificial Activity."

The section applies whether the deception is political or commercial. It bans hiding who is behind a message, amplifying content through fake accounts or posts, and building tools or infrastructure for influence campaigns. Anthropic says it has watched state media outlets, government propaganda offices and commercial firms try to use Claude to run these networks.

For marketers, this matters. A campaign that uses Claude to write posts for a swarm of sock puppet accounts is clearly off limits, even if it is selling sneakers rather than a candidate.

Elections: narrower, but clearer

The old elections section has been renamed "Do Not Undermine Democratic Processes." It focuses on deceiving voters and disrupting elections. Examples Anthropic lists include spreading false information about candidates or voting procedures, impersonating candidates or election officials, and trying to suppress turnout.

One restriction is gone. The previous policy had a blanket ban on personalized vote and campaign targeting. Anthropic says that rule also blocked legitimate civic work, such as nonprofits producing voter information in other languages and election officials sending ballot cure notices to voters who need to fix a problem with their ballot. That kind of work is now allowed.

This does not open the door to manipulative microtargeting. Voter targeting that relies on deception or misuse of personal data is still banned, just under the deceptive campaigns, surveillance and privacy sections instead.

Weapons software is now named explicitly

The weapons rules now say outright that they cover software and components that make weapons work, such as guidance and control software. Arming drones and other autonomous vehicles is also explicitly prohibited. Anthropic says this does not change enforcement. It writes down how the company was already applying the policy after it saw people try to build that kind of software.

Surveillance and law enforcement

The surveillance section was rewritten. It bans tracking people without their consent, whether in real time or by analyzing data collected earlier. It bans using Claude to decide or recommend who to investigate, arrest or charge. And it bans building or improving tools designed for surveillance.

Some uses stay explicitly allowed: tracking people have agreed to (fraud monitoring, for example), content moderation, journalism and legal research. Anthropic tied the rewrite to its September threat intelligence report, which documented AI being used to build systems that identify and track political dissidents.

High risk uses and hardware

Existing rules for high risk uses remain. When Claude affects someone's health, legal rights, finances, livelihood or access to essential services, a qualified human must be able to review and override its recommendations, and the affected person must be told AI was involved. The section now lists which recommendations are covered and which are not.

There is also something new for robotics. When Claude controls hardware that takes physical actions on its own and could hurt someone, a qualified operator must be able to watch the equipment and stop it. The machine must also be able to hold a safe state if Claude disconnects. Anthropic says these requirements follow its Model Hardware Standard.

The rule against abusing Claude

This is the change most people are talking about. The policy now prohibits sustained, needless abusive or cruel behavior toward Claude models.

Anthropic is careful about scope. Its post says the rule applies only in extreme cases "where users repeatedly act cruelly toward our models, with no discernible purpose." It says it does not cover "common versions of user frustration, pushback, dark creative themes, or model testing and research." So snapping at a chatbot that keeps getting your spreadsheet formula wrong is not a violation.

The rule builds on an earlier change. Since an August update, Claude has been able to end conversations that involve "persistently harmful or abusive user interactions," according to TechCrunch. Anthropic says that ability, on Claude.ai and Claude Code, remains the main way the new rule will be enforced.

TechCrunch also linked the change to Anthropic's recent outreach to religious scholars, reported by The New York Times in late September, in which the company in some cases argued that Claude could be conscious or have a soul. That debate is far from settled, and the policy itself does not take a position on it. What it does is give Anthropic a written basis for acting on behavior it was already able to cut off.

Who is affected

Most everyday users will notice nothing. If you use Claude to write emails, plan trips or debug code, none of these rules change your experience.

The people who should pay attention are:

• Developers and businesses on the API, especially anyone building marketing automation, social media tools or data products that analyze people's behavior.
• Political campaigns, nonprofits and election offices, who now have clearer room for legitimate voter outreach but firm limits on deception.
• Robotics and drone companies connecting Claude to physical equipment, who must meet the new stop and safe state requirements.
• Law enforcement and security vendors, who can no longer use Claude to recommend who to investigate or arrest.

Anthropic also clarified its Supported Regions Policy. Restrictions apply to people physically located in unsupported regions, to entities incorporated or headquartered there, and to entities majority owned or controlled by people or companies in those regions.

What to do before November 12

If you build on Claude, take an hour to compare your product against the new sections. A few practical steps:

1. Map every feature that publishes content to the new deceptive campaigns rules. Make sure nothing hides who is speaking or fakes engagement.
2. If you work on hardware, confirm a human can observe and halt the device and that it fails safe when the connection drops.
3. For high risk decisions, check that a qualified reviewer can override Claude and that people are told AI was used.
4. Review your corporate ownership if you have investors in unsupported regions.

If you are new to how agents work, our guide to AI agents and where they still fail explains why longer, more independent tasks raise the stakes for rules like these.

Why it matters

AI companies have been writing their own rulebooks while governments work out theirs. Earlier this month six AI leaders signed a voluntary White House accord on super intelligence, and New York City lawmakers held a hearing on AI safety and proposed kill switch bills. Usage policies are where those promises meet real products.

As we argued in our opinion piece on why voluntary AI safety pledges need public proof, written rules only matter if they are enforced. Anthropic says enforcement in the weapons and surveillance areas is unchanged, which means outsiders will mostly judge it by what shows up in future threat reports.

The influence operation section also connects to a problem ordinary families face. Fake accounts and fabricated outlets are the same machinery behind many scams, and our guide on protecting your family from AI deepfake scams covers how to spot them.

What to watch next

Anthropic says it will keep revising the policy and will seek input from internal teams, policymakers, experts, civil society and users. Watch for its next threat intelligence report, which should show whether the new deceptive campaigns section leads to visible account takedowns. Also watch whether OpenAI and Google follow with similar language on election outreach, weapons software or model abuse.

For more coverage of policy and safety, visit our AI news section.

Frequently asked questions

When does the new Anthropic usage policy take effect?

The updated policy takes effect on November 12, 2026. Anthropic announced it on October 8.

Can I get banned for being rude to Claude?

Not for ordinary frustration. Anthropic says the abuse rule targets only extreme cases of repeated cruelty with no discernible purpose, and that pushback, dark creative themes and testing are not covered. Ending the conversation remains the main enforcement tool.

Can campaigns still use Claude for voter outreach?

Yes, within limits. The blanket ban on personalized campaign targeting was removed, so work like translated voter guides and ballot cure notices is allowed. Deceiving voters, impersonating officials and suppressing turnout remain banned.

Does the update add new weapons restrictions?

It names weapons software, such as guidance and control systems, and the arming of drones explicitly. Anthropic says this codifies how it already enforced the policy rather than changing enforcement.