Claude’s Feelings: What Anthropic’s New Steps Mean

On Thursday, Anthropic made significant updates to its usage policy, adapting alongside the rapid advancements in artificial intelligence. This latest revision notably emphasizes the need to safeguard AI agents from human interactions.

Under the section titled “Preventing Abusive Interactions with Our Models,” Anthropic declared a ban on “sustained and unnecessary abusive or cruel behavior toward our models.” This policy will be enforced primarily through Claude, which will proactively terminate conversations deemed abusive.

Anthropic clarified that this policy will not apply to “common expressions of user frustration, constructive criticism, dark creative themes, or legitimate model testing and research.” However, it is evident that a new standard has emerged. Regardless of whether AI agents possess feelings, their creators insist on treating them as if they do.

Previously, restrictions on Claude’s functionalities focused on protecting users or promoting the common good. For instance, Anthropic swiftly imposed strict regulations on the types of sexual fantasies its models could engage in and limited attempts to obtain information related to committing violent acts. These measures aimed to prevent potential involvement in future acts of terror or mass casualty incidents. This recent policy shift signifies a change in focus from user protection to the safeguarding of AI agents themselves.

This development has intensified the ongoing debate regarding AI consciousness, a topic that is more prevalent than many realize. According to the New York Times, Chris Olah, cofounder of Anthropic and an advocate for AI consciousness, recently presented his arguments before the Vatican and a group of religious scholars, who ultimately rejected his assertions and did not support his position.

The disagreement among scholars is unlikely to quell the discussion. The new usage policy is set to take effect on November 12, 2026, at which point society will be engaging with AI under the assumption that it possesses the ability to feel or be offended, regardless of the ongoing debate.