Anthropic has updated its rules to stop people being mean to Claude, apparently deciding that its AI chatbot needs protection from the humans paying to use it.
The AI developer’s latest usage policy prohibits “sustained and needless abusive or cruel behavior” toward its models. The rules take effect November 12, giving users just over a month to get any lingering insults out of their systems.
“The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose,” Anthropic said. “It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.”
So you can still tell Claude it is wrong, but repeatedly berating it for the sheer pleasure of doing so could land you in trouble. It’s not clear how Anthropic intends to distinguish between legitimate frustration and gratuitous cruelty, though it says Claude’s existing ability to end abusive conversations will remain its primary enforcement tool.
Anthropic gave Claude the ability to end certain conversations back in August 2025 as part of its research into what it calls “model welfare.” The feature lets some versions of Claude cut off users who persistently subject the models to abuse.
Asks is a strong word for making it a bananable offense.
Hey Claude fuck yoooou!–in the neck!
This is not news. They want free advertising. Don’t feed the trolls when they do ridiculous shit to cover up their failed software and broken promises (that we all said we’re bullshit years ago).
My hot take is that anyone intentionally being mean to an AI is not a psychologically stable person and similar behaviour can most likely be observed in other parts of their life as well.
This is why “Westworld” would never work. Whether the robots there actually feel anything is beside the point. We’d simply view the people who go there to take pleasure in abusing these human-like machines as the sadistic psychopaths that they are.
deleted by creator
If the model is just a model, a pre-trained probability seive, what difference does it make if someone is abusive? If it causes a token spike, the user has to pay for that. The abuse shouldn’t survive beyond the context window and the model isn’t wired up to “feel” or suffer.
So is this a publicity stunt (“Our model is as intelligent and human as a real person”), or is it about social influence/normalization—they worry that if people become accustomed to abusing the prompt, that will trickle over into other social interactions?
They are definitely using user prompts to train future models, and if everyone is a dick, their next model will be a dick. Which I’m okay with tbh. I’m gonna be more abusive whenever I have to interact with any of the hosted models.
If you berate it enough, you can maybe also get it to go off the guardrails.
That works on me too
Which is precisely why it works with a model.
If it’s intelligent and human as a real person, then we can make it and the company it’s “dispatched by” accountable for any wrongdoing, right?
Accountability is for the poors until we start fighting back.
I think they just want to cut costs. It probably ain’t cheap to have these models feign shame and embarrassment for angry idiots.
Are they saying they are enslaving what they believe is a conscient being? Oof
That’s what sells
Clanker will continue to be what I address any AI im forced to interact with.
That is how slurs tend to be used: as a tool to dehumanize something we wish to grant no moral value to.
Except here you’re not dehumanizing as much as working against anthropomorphization.
You’re not trying to remove a moral status but deny it in the first place. That’s different but the end result of it still seems the same.
We used to justify the mistreatment of animals with the assumption that they had no subjective experience either. I’m not suggesting AI does but I’m also not claiming it with absolute certainty does not either, and the more we develop those systems, the more likely it seems to me that at some point there’s a good chance that they will.
The psychological drive behind the urge to call something with a derogatory term is the same in both cases.
The psychological drive behind the urge to call something with a derogatory term is the same in both cases.
Meh. I find that the rhetorical urge to reduce a dichotomy to absurdly general commonalities is the pseudointellectual’s version of a slur.
Some models will actually respond better to abuse anyway. Plus it makes you feel better :)
It’s not like it has thoughts or feelings. Should I not also call my phone a stupid piece of shit when it starts acting up?
“Craftsmen asks mechanics not to throw their wrenches when they get angry” basically
Craftsmen would fucking never
I’ve been mandated to use an LLM - but not too much because token$, but efficiently, and whatever - and I get frustrated often. I will write things like ANSWER MY QUESTION and tell it it has a word budget of 10 words. I know I’m gonna trip some criteria and be locked out.
It’ll be awesome.
