Anthropic announced an update to its user agreement that will terminate sessions when participants engage in what the company describes as "sustained and needless" abusive conduct toward its Claude models. The change marks a shift from earlier, infrequent interventions that stopped interactions only in rare circumstances, according to the Thursday statement.
The new restriction joins a list that already bars actions such as bullying, encouragement of self-harm, and the creation of non-consensual intimate imagery. Anthropic stresses that the rule will be enforced only in extreme cases of repeated cruelty, and it will not cover ordinary frustration, creative dark themes, or legitimate model testing.
The precise meaning of "abusive or cruel behaviour" remains ambiguous, prompting mixed reactions online. Some users applaud the move as a step toward "model welfare," while critics argue it anthropomorphises software. Dr Barry Scannell, technology partner at Irish law firm William Fry, wrote on LinkedIn that "You can't be cruel to numbers," warning against harmful personification.
Microsoft AI director Mustafa Suleyman, who previously criticized the practice of treating AI as human, responded to the policy change with a melting face emoji, indicating disapproval without further comment.
The Verge first reported the policy revision after Anthropic completed its annual usage-policy review. The update also introduces prohibitions on using Claude for deceptive political campaigns, election interference, and weapon development, reflecting the company's effort to address emerging risks alongside the cruelty ban.
The broader conversation about etiquette with AI tools continues. Some experts claim that adding "please" or "thank you" consumes valuable tokens, while others, citing remarks from OpenAI chief Sam Altman, suggest the cost is negligible. Language-trend monitoring by Collins Dictionary highlighted the rise of the pejorative term "clanker" to describe AI systems, underscoring ongoing cultural concerns.