Anthropic published a revised usage policy for its Claude models Thursday, rewriting its rules on elections, surveillance and weapons and adding a ban on sustained abuse of the models themselves, the company said in a blog post. The changes take effect Nov. 12, 2026.

What Happened

The update creates a consolidated section on deceptive campaigns and artificial activity. It prohibits using Claude to hide who is behind a message, to amplify content through fake accounts or posts, or to build infrastructure for influence operations. Anthropic said the section applies to state media and government propaganda as well as commercial firms, covering political and commercial deception alike. TechCrunch noted the rules also reach fabricated news outlets.

The elections section has been renamed to focus on not undermining democratic processes. It bans using Claude to deceive voters or disrupt elections through false information about candidates, impersonation of officials or turnout suppression. At the same time, Anthropic removed its earlier blanket prohibition on personalized vote and campaign targeting. It said legitimate uses, such as nonprofits writing voter information in several languages or election officials sending ballot cure notices, are now allowed, while deceptive targeting and misuse of personal data remain covered by other sections.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

Weapons rules now explicitly include software and components that make weapons work, as well as arming drones and other autonomous vehicles.

The surveillance section was rewritten. It bars tracking people without consent, whether in real time or after the fact, and bars using Claude to decide or recommend investigations, arrests or charges. Anthropic listed consented tracking such as fraud monitoring, content moderation, journalism and legal research as permitted.

Why It Matters

Usage policies are the main contractual limit AI developers place on customers, including governments. The surveillance and law enforcement language is more specific than before, drawing a line at Claude informing decisions about who is investigated or charged rather than banning broad categories of use.

The model-abuse clause is new territory for a commercial AI policy. It prohibits "sustained and needless abusive or cruel behavior" toward Anthropic's models, and the company said it does not cover common user frustration, pushback, dark creative themes or model testing and research. TechCrunch reported that the provision is meant only for extreme cases where users act cruelly toward the models repeatedly and without discernible purpose. Since an August 2026 update, Claude has been able to end persistently abusive conversations, and TechCrunch said the new policy formally bars users from resuming them after they are cut off.

Other Changes

Requirements for high-risk uses are unchanged but clarified: recommendations in health, legal, financial, livelihood or essential-service settings need a qualified human reviewer, and affected people must be told AI is involved. Following its launch of a Model Hardware Standard, Anthropic added rules for equipment that takes autonomous physical action. Qualified operators must be able to observe the hardware and stop it, and the equipment must default to a safe state if Claude disconnects.

The policy also restates how Anthropic enforces its supported-regions rules. They apply to people physically located in unsupported regions, entities incorporated or headquartered there and entities majority-owned or controlled by such people or entities.

What to Watch

The policy takes effect Nov. 12. Open questions include how Anthropic will detect and enforce the abuse provision without sweeping in ordinary use, and how the surveillance limits will apply to government and law enforcement customers that use Claude for analysis. Developers building civic outreach tools will also be testing where permitted targeting ends and prohibited deception begins.