WORLD - Anthropic has updated its usage policy to allow its AI tools to end conversations with users who engage in “sustained and needless” abusive behaviour, reigniting debate over how people should interact with artificial intelligence.
The company behind Claude said on Thursday that the change formalises a practice it had previously applied in rare cases. The restriction will apply only in extreme instances of repeated abuse, excluding ordinary frustration, criticism, dark creative themes, and model testing or research.
Policy Defines Prohibited Conduct
The updated policy adds abusive or cruel behaviour toward AI models to a list of prohibited activities that includes bullying, promoting self-harm, and creating non-consensual intimate imagery.
However, Anthropic has not clearly defined what constitutes abusive or cruel behaviour toward its AI systems, leaving questions about how the policy will be enforced.
The changes, first reported by technology outlet The Verge, followed the company's annual review of its usage policy.
Anthropic also introduced restrictions addressing other risks, including the use of Claude for deceptive campaigns and efforts to deceive voters or disrupt elections ahead of the US midterm elections. The company reiterated its existing prohibition on using its tools to develop weapons.
Debate Over AI Treatment
The policy has drawn mixed reactions on social media, with some users welcoming the move as a step toward encouraging respectful interactions. Others questioned its implications and warned against treating AI systems as if they were human.
Barry Scannell, a technology partner at Irish law firm William Fry, criticised the approach, arguing that attributing human-like qualities to AI could mislead people about the technology's nature. Microsoft AI director Mustafa Suleyman, who has previously criticised Anthropic for treating AI as human, also reacted to the announcement.
The update has renewed discussion over whether users should be polite to AI chatbots. Some experts argue that expressions such as “please” and “thank you” consume additional processing resources without being necessary, while others suggest that polite prompts can produce better responses.
Researchers have also raised concerns that routinely using abusive language toward AI could influence how people communicate with one another, although the broader effects remain debated.
