SAN FRANCISCO: Anthropic published a revised Claude usage policy on Thursday that prohibits sustained, needless abusive or cruel behaviour toward its artificial intelligence (AI) models.
Claude’s ability to end those exchanges remains the main enforcement mechanism. Anthropic describes the restriction as targeting extreme cases of repeated cruelty without a discernible purpose.
Ordinary frustration and criticism remain permitted. The company excludes dark creative themes, model testing and research from the restriction.
Anthropic had already given Claude the ability to end conversations in 2025, reserving it for rare cases involving persistently harmful or abusive interactions. The revision adds an explicit prohibition to its user rules.
Its policy does not explicitly invoke model welfare, the idea that AI systems might warrant protections usually associated with living beings. Company executives have nevertheless considered whether such protections could be appropriate.
Anthropic chief executive Dario Amodei told The New York Times in February that he remained uncertain about whether AI models could be conscious. He left open the possibility.
Microsoft AI chief Mustafa Suleyman took the opposite position in a September essay. He argued that AI systems neither experience feelings nor suffer, and warned against granting them rights or moral protections.
Read: Claude Finds Enzyme System in Virus DNA, Anthropic Says
Jackson Stakeman, a general manager at Atlanta-based AI services provider Sparq, questioned whether consciousness was a useful basis for the debate.
He described AI systems as mirrors that reproduce human input at scale, arguing that this was reason enough for Anthropic’s change. Anthropic did not immediately respond to AFP’s request for further comment.
The revised policy takes effect on November 12.
