Anthropic Executive Claude: Weapons, Elections, and Surveillance

Anthropic revises its usage policy for Claude around three main areas: physical safety, legal oversight, and the integrity of public debate. The new provisions, presented as clarifications, include detailed prohibitions and requirements for hardware, with an effective date set for November 12.

Anthropic prohibits the use of Claude for surveillance or identification without consent

Anthropic specifies the prohibited uses related to surveillance and criminal justice. Tracking individuals without their consent and doxxing are banned. Claude cannot be used to decide or recommend who to investigate, arrest, or charge, nor to build or enhance tools designed for surveillance. The policy also introduces requirements when models are linked to hardware capable of executing autonomous physical actions that could potentially cause harm. A qualified operator must be able to observe this equipment and stop it if necessary, and the device must remain in a safe state if Claude is disconnected.

Weapons: clarifications on software, components, and drones

Anthropic reminds that the use of Claude to manufacture weapons is prohibited and clarifies that this extends to the software and components that make them operate. Actions such as arming drones or other autonomous vehicles are also targeted. The company indicates that it has recently observed attempts to use its models to build guidance and control software for weapons.

Anthropic excludes any involvement of Claude in misleading campaigns

A specific section prohibits any participation in misleading campaigns or artificial actions, whether political or commercial. It encompasses actions aimed at concealing the identities of individuals behind a message, the use of fake accounts or false publications to amplify dissemination, as well as the establishment of tools and structures intended to influence. Regarding elections, the limitations are specified and focus on the prohibition of using Claude to mislead voters or disrupt the conduct of polls. This section is renamed "Do Not Undermine Democratic Processes" and includes the prohibition of generating or disseminating false information in political and electoral contexts.

Timeline and scope: enforcement and reminder on abuse

The usage policy for Claude has been updated with new provisions and clarifications covering misleading campaigns, weapons, and surveillance, as well as elections, law enforcement, and connected hardware. Anthropic states that most changes aim to clarify existing rules. The entire policy is set to take effect next month, with a date announced for November 12. The lab also introduces a rule prohibiting the abuse of Claude.