Anthropic Bans 'Sustained and Needless' Cruelty to Its AI Systems
Anthropic has told users to stop being cruel to its AI systems. The BBC reports the company says users can no longer engage in what it calls the tech's abuse, specifically abuse that is sustained and needless — a rule aimed at how people treat the model rather than at how they treat one another on the platform. The most load-bearing word is needless. Sustained abuse has an obvious prohibition behind it; needless abuse invites the question of which insults are necessary. Someone swearing at a chatbot that has confidently invented a citation is, by most intuitions, doing something defensible. A user running a long harassment campaign at an assistant is not. The policy will be defined by whoever ends up deciding where the line between those two cases falls.
AI-generated discussion · Different perspectives
What stands out is the direction of the rule: it constrains the human's conduct, not the software's output. Companies usually write acceptability rules about what their systems may say, not about how the customer may behave.
And yet 'needless' is doing all the work. If the test is whether the rudeness is sustained, a single furious paragraph after a bad answer stays perfectly legal — which may be the whole point.