Anthropic tightens Claude AI rules on model abuse, elections, weapons and surveillance

Anthropic has updated its usage policy for Claude, clarifying restrictions on deceptive political campaigns, weapons-related activities and surveillance, while introducing a new rule against extreme cases of sustained abuse directed at its AI models. The October 8 update comes as Claude handles longer and more independent tasks, prompting the company to explain how existing safeguards apply to newer capabilities.

Not every change introduces a fresh restriction, some provisions make existing rules clearer, while others allow legitimate activities previously covered by broader prohibitions. The revised policy also addresses AI-assisted decisions affecting people's health, finances and legal rights, along with systems that can take physical action.

Must Read: Meta Muse AI is learning more about you than you may realise: Here’s what it tracks

Anthropic changes Claude AI rules

Anthropic has consolidated restrictions on deceptive campaigns into one section. Users cannot employ Claude to operate fake accounts, fabricate news websites, conceal the source of messages or build infrastructure for political or commercial influence operations.

The election rules now focus on preventing voter deception and disruption, including spreading false information, impersonating candidates or election officials, and suppressing turnout. Anthropic has removed its blanket ban on personalised voter and campaign targeting, allowing legitimate activities such as preparing voter information in different languages. Deceptive targeting and misuse of personal data remain prohibited.

Must Read: ChatGPT for teens may not be as safe as parents think: Study finds

The weapons policy explicitly covers guidance and control software, components that make weapons work, and activities such as arming drones. Surveillance restrictions prohibit tracking people without consent, using Claude to recommend whom law enforcement should investigate or charge, and building or improving surveillance tools. Consent-based tracking, fraud monitoring, content moderation, journalism and legal research remain permitted.

Human oversight, abusive behaviour and access restrictions

For decisions affecting health, finances, legal rights, livelihoods or essential services, Anthropic requires qualified human oversight and disclosure to affected individuals when AI is used. For potentially dangerous autonomous physical equipment, operators must be able to monitor and stop it, and it must remain safe if Claude disconnects.

The company has also prohibited sustained, needless cruelty towards its models in extreme cases. Ordinary frustration, criticism, dark creative writing, testing and research are excluded. Claude may end persistently abusive conversations.

The revised policy takes effect on November 12, 2026. Anthropic has also clarified that Claude is unavailable to people in unsupported regions and certain entities incorporated, headquartered, majority-owned or controlled there.