Anthropic bans users from insulting Claude and using AI to develop weapons
Anthropic introduces new restrictions for AI (photo: Planet Volumes)
Anthropic has updated its usage policy for its Claude AI models. The new document clarifies restrictions on using the technology for weapons development, surveillance, and disinformation, and introduces a first-of-its-kind ban on prolonged abuse of the AI models themselves.
Protecting AI from user abuse
The developers are introducing a ban on prolonged and unjustified cruel or abusive behavior toward their models.
The rule will apply only in extreme cases, when users repeatedly act aggressively without any research or practical purpose.
Ordinary expressions of frustration, arguments, model testing, and research are not covered by the restriction. In cases of systematic abuse, Claude will be allowed to end conversations with users on Claude.ai and Claude Code at its own discretion.
Ban on disinformation and manipulation
The developers have consolidated their rules against deception into a new section that prohibits participation in disinformation campaigns and the creation of inauthentic activity.
The move follows cases in which state media outlets, propaganda organizations, and private companies used Claude to operate bot networks and spread fake news on websites.
The ban covers:
-
Concealing the true authors of messages;
-
Artificially boosting content through bot accounts;
-
Creating infrastructure for information operations.
The section on elections has also been updated. Claude may not be used to mislead voters, spread false information about candidates, or disrupt voting.
At the same time, the company has lifted its blanket ban on personalized targeting for legitimate civic initiatives, such as translating voter instructions into other languages.
Restrictions on weapons development and surveillance
Anthropic's policy explicitly states that the ban on weapons development also applies to software and components that enable weapons to function.
Using AI to equip drones and other autonomous vehicles with weapons is also prohibited.
The policy introduces the following rules for surveillance and criminal justice:
-
Surveillance of people without their consent is prohibited, whether conducted in real time or through the analysis of collected data;
-
AI may not be used to make decisions about investigations, arrests, or the filing of criminal charges;
-
Developing or improving surveillance tools is prohibited.
However, fraud monitoring with users' consent, content moderation, journalism, and legal research remain permitted.
Requirements for high-risk decisions
When AI is used in healthcare, finance, or law, a qualified professional must be available to review and override the system's recommendations. Users must also be informed that AI is being used.
If AI models are connected to physical equipment capable of causing injury, the operator must be able to monitor the machinery and stop it when necessary.
If the connection to the AI system is lost, the equipment must automatically switch to a safe mode.
The updated policy will take effect on November 12, 2026.
