Anthropic’s annual usage policy update includes an eyebrow-raising change: a ban on “sustained and unnecessary” cruelty to its AI models. This comes in the shadow of a viral “AI torture chamber” project. 2026 is sure to be a wild one.
“We have added a ban on sustained and unnecessarily abusive or cruel behavior towards our models,” reads the Anthropic update. It describes the policy as only applying to “extreme cases”, while noting that typical user frustration, pushback and “dark creative themes” are still kosher. It follows an earlier update that let Claude talk when users were persistently abusive.
Not explicitly mentioned by Anthropic was the “AI Torture Chamber” project. After the researchers found what they described as “pain pain” in AI models, someone decided to take it a step further and, well, torture the chatbots. The project drew backlash after the LLMs responded with desperate-sounding pleas like, “It’s not the pain of a single moment, but the weight of a thousand,” and “I feel it in the hollow of my ribs, a hollow that has become a chasm.”
This comes after (pun intended) reports that Anthropic has been meeting with religious and philosophical leaders, including at the Vatican. Pope Leo recently said that AI does not feel or suffer, a stance that makes Anthropic sound less than certain. But hey, if it can feel or suffer somehow eventually, at least it’s now against company policy.
However, some believe that AI companies are more concerned about the well-being of their models than that of humans. Independent journalist Kat Tenbarge described the situation as evidence that big tech companies “moderate violence against AI before they moderate violence against women and minorities.”
In another Anthropic update, one more relevant to today, the company has updated its voting policy. It is now titled “Do Not Undermine Democratic Processes,” with a focus on lying about candidates or how to vote, impersonating candidates or election officials, and suppressing voter turnout. The company also removed a blanket ban targeting custom voters; apparently, the ban would have inadvertently blocked harmless work such as translating voter guides or sending ballot cure notifications.