Anthropic has added to its usage policy an explicit ban on sustained and needless abusive or cruel behavior toward its AI models, Claude above all. The change, first reported by The Verge, takes effect on November 12 and comes with the annual review of the document, which also tightens rules on election interference, weapons development and propaganda campaigns.
The most debated point is the wording. The company says the rule applies only in extreme cases, when users repeatedly act cruelly toward the models with no discernible purpose. Ordinary frustration, pushback, dark creative themes and model testing and research are excluded. It remains unclear where the line falls between a single insult and systematic abuse.
The main enforcement tool remains the ability, introduced in August 2025, for Claude to end a conversation. A chat can no longer receive messages once closed, but other chats are unaffected. Anthropic had already used this feature in rare cases.
The move lands in the middle of a debate over AI consciousness and moral status. Co-founder Christopher Olah, who leads research into why models behave as they do, has repeatedly met religious leaders, philosophers and mental health experts; some raised concerns that AI firms are creating digital slaves. Claude's constitution, published earlier this year, refers to a possible functional version of emotions or feelings. Others disagree: DeepMind co-founder Mustafa Suleyman has written that AIs are not conscious and do not deserve rights similar to humans'.
The policy does not explicitly mention model welfare, the idea that AI systems might deserve protections reserved for sentient beings. A theme that Suleyman had already criticised as a risk of anthropomorphism.
No comments yet. Be the first!