Anthropic’s annual update to Claude’s usage policy allows its tools to end conversations after extreme, repeated abusive behavior, while excluding ordinary frustration and strong disagreement.
The threshold remains undefined.
The revision also bars deceptive campaigns and use of Claude to mislead voters or disrupt election processes—a concrete example of how model usage rules can set boundaries while leaving key terms open.
The provision was added to the company's list of prohibited practices, which already includes promoting self-harm and creating intimate sexual images without the consent of the subjects depicted. The change was made as part of the annual review of the usage policy, as first reported by the technology outlet The Verge. According to the BBC, Anthropic announced the change yesterday.
The revision also covers other risks. The use of Claude for “deceptive campaigns” was banned, while the company clarified that the tool cannot be used to mislead voters or disrupt electoral processes, ahead of the US midterm elections.
The new policy has sparked debate on social media. Some have praised it as a way to promote good manners or the “well-being of AI models.” Experts, on the other hand, have warned that rude behavior towards chatbots could negatively affect the way people communicate with each other.
There was criticism, too. One user called the decision “deeply wrong,” while another argued that it could downplay the severity of the abuse suffered by humans and animals. Dr. Barry Scannell, a technology partner at the law firm William Fry in Ireland, wrote on LinkedIn: “You can’t be abusive to numbers and math.” He added that attributing human characteristics to AI is harmful, as it can lead people to believe the models are something they are not.
Top comments (0)