Anthropic has published an updated Usage Policy for Claude that will take effect on November 12, citing new AI capabilities, customer feedback and misuse cases identified over the past year. The update bans the use of Claude to create surveillance tools or enable weapons operation, while setting more specific safeguards for high-risk uses and equipment capable of physical actions.

Anthropic said most of the updates clarify existing rules rather than change how the company enforces them. The revisions include restrictions related to disinformation, weapons and surveillance, along with more detailed requirements for uses that could affect people’s health, finances and safety.

Disinformation and election rules

Anthropic consolidated earlier restrictions on deceptive conduct into a separate section called “Do not engage in deceptive campaigns or inauthentic activity.” The rules apply to political and commercial campaigns.

The policy prohibits concealing who is behind messages, artificially amplifying content through fake accounts and posts, and creating tools or infrastructure for influence operations. Anthropic said those restrictions previously appeared across sections covering elections, fraud, privacy and disinformation.

The company also renamed its elections section “Do not undermine democratic processes.” It prohibits using Claude to mislead voters or disrupt elections, including by spreading false information about candidates or voting procedures, impersonating candidates or election officials, or attempting to suppress turnout.

Anthropic removed its absolute ban on personalized targeting of voters and campaigns, saying the previous language could encompass legitimate activities such as preparing voter information in different languages. Targeting based on deception or misuse of personal data remains prohibited.

Weapons and surveillance

The revised weapons rules explicitly cover software and components that enable weapons to function, including targeting and control systems. They also prohibit using Claude to arm drones and other autonomous vehicles.

Anthropic rewrote its surveillance and criminal-justice provisions after noting increased use of AI to build systems that identify and track political dissidents. The updated policy prohibits tracking people without their consent, both in real time and through analysis of previously collected data.

Claude also may not be used to make decisions or recommendations about whom law enforcement authorities should investigate, detain or charge in criminal proceedings. The policy further prohibits creating or improving tools designed for surveillance.

Permitted uses include consent-based tracking, such as for fraud detection, as well as content moderation, journalism and legal research.

Requirements for high-risk uses

Anthropic further clarified its rules for uses in which Claude’s recommendations may affect health, legal rights, finances, livelihoods or access to vital services. A qualified person must be involved to verify the model’s recommendations and change them if necessary, and the person affected by the decision must be informed that AI was used.

Those requirements existed previously, but the updated policy describes more specifically which recommendations they cover. It also sets requirements for using Claude to control equipment capable of performing physical actions.

Abusive treatment of models

The policy adds a ban on prolonged and unwarranted cruel or abusive behavior toward Anthropic’s models. The company said the rule is intended to apply only in exceptional cases in which a user systematically abuses a model without a clear purpose.

The restriction does not cover ordinary dissatisfaction, arguments, dark creative themes, testing or model research. Anthropic said the rule is consistent with Claude’s existing ability to end certain conversations on Claude.ai and Claude Code when users systematically engage in abuse.

Anthropic previously launched a free vulnerability scanner for open-source software.

Source: Incrypted