New Delhi: Anthropic has announced a revised usage policy that prohibits sustained, purposeless abuse of its Claude artificial intelligence models while clarifying restrictions on weapons development, surveillance and deceptive campaigns. Announced on Thursday, the changes take effect on November 12 and distinguish extreme abusive conduct from ordinary criticism, frustration and legitimate research.
The provision attracting particular attention bars “sustained and needless abusive or cruel behavior” towards the company’s models. However, Anthropic’s explanation of the update says the restriction targets extreme, repeated cruelty without a discernible purpose, while allowing users to challenge Claude’s answers, explore dark fictional themes and conduct model testing.
That distinction matters for users concerned that expressing annoyance could cost them access to the chatbot. As the Verge reported, Claude’s ability to end persistently abusive conversations will remain the principal enforcement mechanism for this provision.
Account restrictions nevertheless remain available under the company’s broader usage policy, which permits warnings, reduced access, suspension or termination when Anthropic suspects a violation. The document also says that an automated safeguard blocking an output does not, by itself, establish that the user has broken its rules.
The conversation-ending feature predates the latest announcement: Anthropic introduced it for Claude Opus 4 and 4.1 on August 15, 2025. Its original description characterized the feature as a last resort after repeated attempts to redirect a harmful exchange had failed, with additional protections for users at imminent risk of harming themselves or others.
Ending an exchange also differs from disabling an account. Anthropic said users could immediately start another chat, continue using other conversations, or edit earlier messages to create a new branch of an ended discussion.
Beyond chatbot interactions, the revision consolidates restrictions on deceptive political and commercial activity into a dedicated section. Anthropic says it has observed state media organizations, government propaganda offices and commercial businesses using Claude to operate fake accounts and fabricated news websites, helping explain the emphasis on concealed sponsorship and misleading attribution.
The election provisions address deceptive voter communications, impersonation and interference with voting, alongside false information intended to suppress participation. Comparing the previous policy with the revision shows that many underlying prohibitions already existed, although their organization and wording have changed.
The weapons restrictions now explicitly cover software and components used in targeting, fire control and weapons operation, as well as arming drones and other unmanned platforms. Surveillance provisions prohibit tracking people without consent and developing tools for prohibited surveillance, while barring Claude from making or suggesting criminal justice decisions such as whom to investigate, arrest or prosecute.
The update also addresses AI systems connected to potentially dangerous physical equipment. A qualified person must be able to observe and stop such equipment, which must enter a safe condition when that person intervenes or the connection to Anthropic’s services fails.
For high-risk advice and decisions affecting individuals, including medical, legal and financial applications, the policy requires meaningful review by a qualified person and disclosure of AI involvement. General educational explanations and internal research that do not themselves deliver such recommendations are among the exclusions.
The unusual restriction on cruelty towards Claude sits alongside Anthropic’s longer-running research into potential AI welfare, announced on April 24, 2025. The company has acknowledged substantial uncertainty about whether present or future AI systems could be conscious or have experiences warranting moral consideration, and says there is no scientific consensus on those questions.
The policy therefore does not establish that Claude has feelings or consciousness. It sets a precautionary boundary around extreme interactions while extending rules governing how increasingly capable AI systems are used to influence people, support decisions and control equipment.