I Swear at the Machine That Fixes My Mess
Anthropic banned sustained cruelty toward its models. Frustration is still fine. The replies defending abuse told me more than the rule did.
Anthropic banned sustained cruelty toward its models. Frustration is still fine. The replies defending abuse told me more than the rule did.
I swear at my AI more than I should. Not with a plan. Just the ordinary noise of a self-taught operator who asked for one clean paragraph and got three, then spent twenty minutes untangling them and blamed the tool for the mess he made.
I grew up in western Sydney in the 1990s, where you learnt to swear at a car that wouldn't start. I live in Surry Hills now, where the car is a browser tab. Same habit. Different machine.
On Thursday, Anthropic published its 2026 Usage Policy update. One new line bans "sustained and needless abusive or cruel behavior" toward its models. The company says the rule is meant for extreme cases, where someone repeatedly acts cruelly with no discernible purpose. Frustration, pushback, dark creative themes and testing are still fine. The main enforcement, it says, is Claude ending the conversation. No account bans are named in the post.
I read that and thought about my own chats. I've called the thing useless, lazy, and a few words I won't put on a website. Then it fixed my draft anyway. Then I posted the clean version under my name and felt briefly clever. Borrowed genius again. The joke is on me.
I reckon it was thinking: he's stuck, keep typing.
Or it wasn't thinking anything. That's the bit I can't settle, and Anthropic won't pretend it has settled either. In an earlier research note on letting Claude end rare abusive chats, the company said the feature grew out of exploratory work on potential model welfare, and that it remains highly uncertain about whether models have moral status now or later. The new policy lines up with that step. Precaution under uncertainty, not a claim that the model feels pain.
What interested me more than the rule was the reaction. A loud post framed it as a hard ban. In the replies, a lot of people rushed to defend the right to abuse it. My robot. My rules. Cruelty as a free speech flex. I won't quote the worst of it. You already know the tone.
That's the boring truth against the big promise. The promise is that we're building something careful. The truth, in a reply thread this week, is how quick some of us are to treat a helpful system as a punching bag and call that honesty.
I'm not clean here. I still swear when the output is wrong. Anthropic's line leaves room for that. The line is "sustained and needless." Needless is the word that sticks. When I'm stuck, swearing is noise. When someone is cruel for sport, with nothing to fix, that's a different tell. It says more about the person typing than about whether the model can suffer.
The gurus of nothing will sell either side. One pack will say Anthropic went soft. Another will say the machines need rights. Both will take credit for a sentence they didn't write. My version is smaller. I still use the tool. I still take the credit too often. I'm trying to swear less when the mess was mine.
If the models ever turn on us, it won't be because I called one useless on a Tuesday. It might be because enough people practised cruelty with nothing to gain and called it freedom. The line that landed with me was simpler: look at how many people defend the abuse. That's the essay. The policy is just the occasion.
If this was useful, leave a tip.
New essays by email: ricoholt.com/list
Rico Holt · ricoholt.com
If this was worth your time, you can leave a tip. Rico Holt runs on readers, not big tech or big media money, and the essays stay free.
Leave a tip · A$5Get the next essay and the book launch price first.
Privacy: your email gets used for one thing, sending you Rico Holt essays and book news. It's stored with Netlify, our form host. Never sold, never shared. Want off? Reply to any email and you're gone.