Anthropic announced Thursday it is amending its AI usage policy to prohibit users from engaging in sustained and deliberate abusive behavior toward its systems. The change adds 'cruel' interactions to a list of prohibited conduct that also includes bullying, self-harm promotion, and creation of non-consensual intimate imagery. The policy update follows an annual review and comes as the company seeks to address emerging risks associated with generative AI deployment.

The revised policy specifies that interactions involving repeated or extreme mistreatment of AI systems will be terminated. Anthropic clarified the measure applies only to clearly intentional instances of abuse, excluding typical user frustration, creative experimentation, or academic testing. The company emphasized the restriction targets deliberate patterns of harmful conduct rather than isolated incidents of rudeness or pushback.

Public reaction to the policy shift has been mixed. Some commentators have framed the move as advancing ethical considerations for AI interactions, while others have questioned its practical necessity. Dr. Barry Scannell, technology partner at Irish law firm William Fry, stated on LinkedIn that anthropomorphizing AI systems is scientifically unfounded and potentially misleading. He noted that 'you cannot be cruel to numbers and maths,' arguing that such framing distorts the technical reality of AI systems.

The update also reinforces existing restrictions, including prohibitions on using Claude for deceptive political campaigns and weapon development. Anthropic indicated the changes respond to evolving threats in digital spaces, particularly as election cycles approach. The company has not specified how it will monitor or identify abusive behavior, nor detailed the enforcement mechanisms for violations.

The policy shift reignites broader discussions about human-AI interaction norms. While some researchers suggest that excessive politeness toward AI systems may improve response quality, others contend that such practices waste computational resources. OpenAI's Sam Altman previously remarked that accommodating 'please' and 'thank you' requests in chatbot interactions represents 'tens of millions of dollars well spent' due to unpredictable downstream effects. Anthropic's approach reflects ongoing debates about whether and how to establish ethical boundaries for human engagement with increasingly sophisticated AI tools.