Anthropic has released an update to its 2026 usage policy, which prohibits continuous and meaningless abuse or cruel behavior towards Claude, effective November 12.

The company specifies that this clause only targets extreme cases: users repeatedly and cruelly treating the model without visible purpose. Normal dissatisfaction, rebuttals, dark-themed creations, model testing, and research are not included. The main enforcement method remains to end the conversation. This aligns with the capability launched in August 2025: Claude can proactively terminate a few conversations that are continuously harmful or abusive on Claude.ai and Claude Code.

The election section has been renamed to prohibit undermining the democratic process, focusing on banning the use of Claude to deceive voters or disrupt elections. Examples include spreading false information about candidates or voting methods, impersonating candidates or election officials, and suppressing voter turnout. Additionally, deceptive advertising and false activities are prohibited, such as operating fake accounts or fabricating news organizations using the model.

The weapons clause has been explicitly expanded to include software and components that operate weapons, as well as arming drones and other autonomous vehicles. The company states that it has seen attempts to use the model for weapon guidance and control software over the past year. The monitoring clause has been changed to prohibit tracking individuals without consent, whether in real-time tracking or analyzing existing data. In threat intelligence reports, the company has recorded instances of Claude being used to identify and track political dissidents.

The update also clarifies the requirements for high-risk uses such as medical and financial applications, and gives the model the autonomy to take physical actions and control: qualified operators must be able to observe the equipment and stop it when necessary, and the equipment must be able to maintain a safe state after disconnection from Claude. The company states that most changes are clarifications of existing rules, based on the model's ability to perform longer and more independent tasks, as well as patterns of misuse over the past year in influencing actions, weapon development, and monitoring.

This is not a transactional event. The buyers are enterprises and developers continuing to use Claude, while the sellers are the uses blocked by the terms. The event is driven by the annual policy text, not by a price reduction of the model. Funding remains in subscription and interface fees, but election propaganda, fake accounts, weapon guidance software, and unauthorized tracking will be rejected. The beneficiaries are compliant customers who need audit records, while those under pressure are projects treating the model as components for propaganda, monitoring, or weapon software. Ending conversations remains the first line of defense, with account bans following repeated violations.

Source: Public Information