Anthropic revises the rules for using Claude models
Anthropic has revised the rules for Claude models: it prohibits sustained, purposeless cruel treatment, consolidates restrictions on deceptive campaigns, and explicitly covers weapons software and surveillance of people without their consent through analysis of previously collected data.
Anthropic has updated the rules for using Claude models for the first time in more than a year. The rules now prohibit sustained and purposeless harsh or cruel treatment of models. According to the company, the ban applies only to extreme cases, not ordinary frustration, disagreement, dark creative themes, or testing and research. Ending the conversation remains the main enforcement mechanism; the company has not commented on whether users might be blocked.
The company has consolidated previously scattered restrictions into a ban on deceptive commercial and political campaigns. The rules restrict concealing who is behind a message and distributing content through fake accounts or posts. They also explicitly prohibit deceiving voters, impersonating candidates or election officials, and attempts to suppress voter turnout.
The ban on weapons development now explicitly includes software and components that enable weapons to function, as well as arming drones and other autonomous vehicles. According to the company, the change responds to repeated attempts to use Claude models to develop guidance and control software for weapons.
The rules prohibit surveillance of people without consent, both in real time and through analysis of previously collected data. Claude models must not be used to select or recommend people for investigation, arrest, or charges as part of law enforcement or criminal justice procedures, or to create and improve tools intended for surveillance. The company allows adjustments to the rules in contracts with certain government customers if, in its assessment, contractual restrictions and safeguards sufficiently mitigate potential risks.
Why it matters
Assessing whether a use is permitted requires more than checking the model output itself. The rules also take into account how messages are distributed, the purpose of personal data analysis, and the role of software in the functioning of weapons. The surveillance restrictions therefore also apply to work with existing data.
What was added since the original report
Verified updates
-
The ban on cruel treatment does not apply to ordinary frustration, disagreement, dark creative themes, or testing and research.; Since the August update, Claude has been trained to end persistently harmful or abusive conversations.
- The ban on cruel treatment does not apply to ordinary frustration, disagreement, dark creative themes, or testing and research.
- Since the August update, Claude has been trained to end persistently harmful or abusive conversations.
Two audiences, two different impacts
What this means
For individuals
According to the company, ordinary disagreement with a response or testing of models does not fall under the new ban on cruel treatment. In cases of repeated, purposeless harsh treatment, the model may end the conversation.
More practical updates →For a business
Companies using Claude models in campaigns or to analyze data about people must assess whether their entire workflow complies with the rules, particularly the use of fake accounts, concealment of who is behind a message, and surveillance of people without consent.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.