Anthropic has updated its usage policy for Claude to ban what it describes as sustained and needless cruelty toward the model, with the revised rules taking effect on 12 November.
The policy change, published on 8 October, stops short of a sweeping overhaul. Anthropic says most of the revisions clarify rules that already existed rather than introduce new enforcement mechanisms. The cruelty clause is explicitly narrow in scope, covering only extreme cases and excluding ordinary complaints, dark creative writing, and model testing or research.
As reported by Hellomarvisaitoday, the clause is directly tied to a feature already live in Claude.ai and Claude Code that allows the model to end a conversation with a user who persists in abusive behaviour. Anthropic says that capability will remain the primary means of enforcement, so the practical change for most users is limited.
The revised policy also reorganises how it handles deceptive activity. Rules previously spread across separate sections covering elections, fraud, privacy and disinformation have been consolidated under a new deceptive-campaigns heading. The elections section has been narrowed to focus specifically on voter deception and interference with electoral processes. Anthropic says the structural changes reflect misuse patterns it documented in a threat intelligence report published in September, which covered influence operations, weapons development and surveillance.
Clarifications on weapons and surveillance are also included, touching on software and components that make weapons operational, the arming of drones, tracking individuals without consent, and recommending investigation or arrest targets. Anthropic is clear that these clarifications do not alter actual enforcement.
Companies and teams building on Claude face a consolidated rulebook as a result. High-risk use cases affecting health, legal rights or finance still require a qualified person authorised to review and correct the model's output, along with disclosure to affected individuals that AI is involved. Because many users had asked whether their specific use qualified, Anthropic has now listed covered and uncovered activities across eleven fields, including legal, medical, financial and employment.
The policy also introduces requirements for equipment that operates autonomously and could cause physical harm, drawing on a Model Hardware Standard Anthropic announced in August. A qualified operator must be able to monitor and halt such a device, and it must default to a safe state if its connection to Claude is interrupted.
The backdrop to all of this is Claude's expanding role in longer, more autonomous tasks over the past year. The updated policy adds examples of how existing rules apply to those newer capabilities, keeping the rulebook current with how the product is actually being used.
