Thought Leadership
The usage policies prohibit sustained abusive behavior towards Claude. In extreme cases, the model can end conversations.
Anthropic has expanded its usage policies to include a prohibition on sustained and needlessly abusive or cruel behavior towards its AI models. This is reported by Cybernews. Following the change, Claude can end such conversations. Anthropic does not give a reason for the adjustment in the policies. The company initially did not respond to an inquiry from Cybernews.
Only for extreme cases
The rule applies only to extreme cases in which users repeatedly treat the model cruelly. Common forms of frustration, pushback or dark creative themes are not meant to be covered by it. Testing and research environments are also exempt.
As early as August 2025, Anthropic had given the models Claude Opus 4 and 4.1 the ability to end conversations in rare cases of persistently abusive interactions. At the time, the company described this as part of its research into the potential welfare of AI models.
Research into AI welfare
In April 2025, Anthropic had launched a research program on so-called model welfare. It examines whether and when the welfare of AI systems should be given moral consideration, what significance preferences and possible signs of distress in models have, and which practical, low-cost measures would be possible. “Should we also be concerned about the potential consciousness and experiences of the models themselves?”, Anthropic asked at the time.
At the same time, the company emphasized that there is no scientific consensus on whether current or future AI systems could be conscious or could have experiences that deserve consideration. There is also no agreement on how these questions can be investigated. Anthropic referred to a report by experts, including the philosopher David Chalmers, according to which AI systems with consciousness and high degrees of agency are possible in the near future and could deserve moral consideration.
(Editorial Team)
