IT之家 · 智能时代 · 10/9/2026, 07:31:17
Anthropic Updates Claude Policy: Bans Model Abuse, Expands Weapons and Surveillance Restrictions
Anthropic announced updated usage policies effective November 2026, introducing a ban on 'persistent and unnecessary abuse' of the model while significantly tightening restrictions on weapons development, autonomous vehicle control, and non-consensual surveillance. The new rules mandate human oversight for AI connected to physical hardware but exempt common complaints, dark creative themes, and research testing.
SOURCE COVERAGEFull Original Report
On October 9, tech media outlet The Verge published a report stating that Anthropic has revised its Claude usage policies, effective November 12, 2026. The update introduces, for the first time, a clause prohibiting "sustained and unnecessary abuse or cruelty." This rule applies to cases where users repeatedly engage in cruel behavior toward the model without clear purpose.
Regarding enforcement mechanisms, Anthropic stated that the primary method is terminating conversations, but did not specify whether further measures such as user bans would be introduced. The company emphasized that this clause does not apply to common complaints, skepticism, dark creative themes, or model testing and research. IT Home attached relevant screenshots below:
Anthropic has expanded its weapons ban to cover software and components that enable weapon operation, as well as armed drones and other autonomous vehicles. The company noted that it has identified multiple attempts to use Claude to develop guidance or control software for weapons.
Surveillance clauses explicitly prohibit tracking individuals without consent, whether through real-time monitoring or analysis of existing data. Claude must not be used to determine or recommend targets for investigation, arrest, or prosecution in law enforcement and criminal justice proceedings, nor to build or improve surveillance tools.
Anthropic also stipulated that when AI is connected to autonomous physical hardware capable of causing harm, qualified operators must monitor the equipment and halt operations if necessary. For specific government clients, the company may adjust these rules if contractual restrictions and security measures are deemed sufficient.