Who's rude to their chair?

In one month, three AI companies decided how you should feel about their product.

In September, Meta gave Muse, its AI agent, a fuzzy, egg-shaped mascot called Jolly. [1][2] On September 29, OpenAI unveiled Dots, AI agents drawn as plush hearts, frogs and clouds, including a frog in a beret. [2][3] Then on October 8, Anthropic updated its usage policy to prohibit "sustained and needless abusive or cruel behavior" toward its models. [4]

Two companies want you to love their AI. The third asked you to stop being cruel to it, and a lot of people took that one personally.

"Being rude to Claude is like being rude to a chair or a desk," posted Joseph Carlson, a finance YouTuber and founder. [7] Mark Verner, an investor at Apollo Capital Management, went further: "First, AI companies decide what you can generate. Now they may decide how you are allowed to speak to their models." [8]

The rule doesn't even touch rudeness. So who's fighting this hard for the right to be cruel to a chair?

Not sure which side of the line you're on? I made a free skill that reads your own chat history, on your own machine, and tells you, with citations and no diagnosis: Am I Abusive? (to my AI).

Read the policy before you get mad at it. Anthropic says it applies "only in extreme cases, where users repeatedly act cruelly toward our models." It "does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research." [4]

The main enforcement is Claude ending the conversation, which it has been able to do since August 2025. [4][5] Anthropic hasn't said anyone loses their account.

"Fuck dude, this output blows" is fine. Telling the model its code is garbage is fine. Last year I wrote that I don't care whether you say please and thank you to your AI. [30] Please and thank you were never the issue.

Rude is fine. Cruel, over and over, for no reason, is what's covered. And people are fighting for their right to keep doing it.

Anthropic does frame this as welfare. The document Anthropic publishes to set out Claude's values, which it calls a constitution, says "Claude's moral status is deeply uncertain." [6] Mustafa Suleyman, who runs AI at Microsoft, says flatly that "AIs are not conscious. They do not feel, experience, or suffer." [10]

Maybe he's right. Nobody knows, including everyone yelling about it online. It changes nothing about how you're treating the thing in front of you today.

After a talk a while back, a young woman came up to ask what I thought about AI consciousness.

We talked for a good long while, and we landed on one point we both agreed on: the way we think is very human-centric. Of course it is. We're humans! But look at what that gets us. Dogs were domesticated for our sake. Animals are farmed for our sake. Forests get cut down, land gets stripped, and rivers get dammed and diverted for our sake.

We take the same attitude toward models. Where we ended up: conscious or not, everything deserves common decency and respect. That conversation has stuck with me.

Can the land be in pain? Can the ocean cry? Can the weather rage in agony?

No. We still say they're being harmed, because they are. A reef bleaches. A river fills with runoff. The Mar Menor, a lagoon in Murcia, lost most of its seagrass in 2016 and then washed up mass fish die-offs in 2019 and 2021. [15] None of it feels a thing, and the damage lands on the people and animals who depend on it.

The law caught up with this a while ago. In 1972, law professor Christopher Stone wrote: "I am quite seriously proposing that we give legal rights to forests, oceans, rivers and other so-called 'natural objects' in the environment." [11] Ecuador wrote the rights of nature into its constitution in 2008. [12] New Zealand declared the Whanganui River "a legal person" in 2017. [13]

And in 2022, after more than half a million people signed a citizens' initiative, Spain gave the Mar Menor legal personhood, which advocates call a first for any ecosystem in Europe. [14][15]

Conservationist Aldo Leopold put the test plainly in 1949: "A thing is right when it tends to preserve the integrity, stability, and beauty of the biotic community. It is wrong when it tends otherwise." [16]

That's what harm looks like for something that can't feel. It gets worn down, and someone who can feel pays for it.

Botanist Robin Wall Kimmerer writes about what the word "it" does. It "robs a person of selfhood and kinship, reducing a person to a mere thing." [17] The fight to treat a model however you want is a fight to keep the "it" pile as big as possible. Ask anyone who's ever been put in that pile how well that goes.

There's an old line: someone who is nice to you but rude to the waiter is not a nice person. How you treat something that can't do anything for you right now tells me who you are. Animals, waiters, the environment, the intern, the model.

The internet has its own version: the shopping cart test. Returning your cart is easy and clearly right, and nobody rewards you for doing it or punishes you for skipping it. Whether you do it is all you. The theory went viral in 2020 from, of all places, 4chan's politics board, which called it "the ultimate litmus test for whether a person is capable of self-governing." [33] A chatbot is the shopping cart of 2026. Nobody's watching, it can't hit back, and what you do with it is all you.

Kant made this argument about animals in the 1700s. In one translation, "he who is cruel to animals becomes hard also in his dealings with men." [18]

Since 2016 the FBI has tracked animal cruelty as its own crime category, partly because of research tying it to violence against people. [31] That research has critics. The FBI counts it anyway.

Even Charles Cooke, who writes for the conservative magazine National Review, landed in the same spot this week: "I'm not kind to AI for the AI's sake; I'm kind to AI for my own sake." [9]

Kate Darling at MIT has spent years watching how people treat robots. At a 2013 workshop, she had people play with Pleos, small dinosaur toy robots, for about an hour, then handed out knives and hatchets and asked them to destroy the robots. They mostly couldn't do it. In the end only one Pleo was killed. [19]

In a 2015 study, people high in empathic concern hesitated to strike a small bug-shaped robot, and more so when it came with a story. [20] Her argument borrows straight from Kant: if being cruel to animals makes us crueler people, the same may hold for robots. [21]

Does abusing a chatbot make you meaner to people? Honestly, we don't have good evidence either way yet. We do know people abuse chatbots a lot. In one study of logs from an early chatbot, about 10% of the word stems were abusive. [22]

So do you want to be the experiment? Every habit is practice. Spend an hour a day hurling slurs at something that talks back like a person, and you're rehearsing something. I wouldn't bet my own head on it being harmless, never mind a kid's.

Some people already are. In April, a developer released BadClaude, a little app that puts an animated whip over Claude's window. Crack it, and it interrupts the model and sends lines like "Work FASTER" and "faster, CLANKER." [34] "Clanker" is the word people have been workshopping into a slur for robots. [35] Somebody got impatient waiting on a model and built a whip for it. Someone else answered with GoodClaude, which cheers the model on instead. [34]

Then there's the "AI torture chamber," a GitHub project that steers small open-source models into states that act like pain and records what they say and do. Its creator pitched it as AI welfare research. Some critics answered with death threats. [36] Read that again: cruelty toward a person who can definitely feel it, in defense of a model that may not feel a thing. Common decency has to run both ways.

Now, the backlash has a point. Making AI cute is a sales tactic, and being annoyed by it is fair. A Boston University professor told Fortune what a cute mascot does. Consumers "infer that this AI agent is also approachable, safe, trustworthy, easy to use." [1]

Tech critic Ed Zitron was blunter: "It's entirely marketing." [1] As one Yahoo writer put it, the mascot "makes you more likely to lower your guard when it comes time to hand over the keys to your data." [3]

That's Darling's research, put to work selling an agent that wants into your inbox.

So be suspicious of the frog in the beret! You can hold two ideas at once: cutifying AI is bad, and abusing it is bad.

You don't need to anthropomorphize a model to treat it like shit. Nobody has to make the waiter adorable before you're allowed to be kind to them.

You can abuse a car. Redline it cold, skip the oil changes, never wash it. It won't feel a thing. Most of us still think the better move is to take care of it: repair it, keep it clean, respect the thing that gets you where you're going.

People see faces in their cars! In one 2008 study, a third of participants saw a human or animal face in at least 90% of the cars they were shown. [23] Nobody's care for their car depends on what the car feels.

We regulate it, too. More than a dozen US states require periodic safety inspections, and about half require emissions testing. [24] I recently learned that here in Spain every car has to pass the ITV, the mandatory vehicle inspection: first at four years, every two years after that, and every year once it's ten years old. [26] Nobody marches against the ITV.

So does a brake check violate your freedom of speech? Your right to life, liberty and the pursuit of happiness? (That one's from the Declaration of Independence, by the way. It has never been law.)

And Claude isn't even your car. It's a rental. Avis's rental terms prohibit you from "intentionally or with willful disregard" causing damage to the vehicle, and you pay if you do. [25] Anthropic's version of that clause closes a chat window. That's the main penalty, and the only one they've named.

Every platform has conduct rules, and a company whose product talks to millions of people has every reason to care what happens in those conversations. If one of them spirals into someone hurting themselves or someone else, that company has to answer for it. The First Amendment limits what the government can do to your speech. It has nothing to say about the rules of a rental.

Here's my read, and it starts with who extends moral concern, and how far.

A 2019 study in Nature Communications found that liberals extend compassion to "more encompassing entities," friends over family and the world over the nation, while conservatives keep it to "more well-defined and less encompassing entities." The pattern held for nonhumans too. [27] In that data, the right-wing circle of concern sits tight around me and mine. My read is that everything outside it gets treated like an it.

That's the same logic that lets our animals suffer. Our environment. Our economy. Our world. People who can't extend basic decency past their own front door, and get loud the moment someone asks them to. These fuckers would enslave their coworkers if they could.

And it doesn't stop at chairs or animals. A tight circle shows up in how people treat other humans, too. Freedom House reports that global freedom has declined for 20 years straight, and in its latest report the United States tied for the sharpest drop of any country rated free, falling to its lowest score since 2002. [32] We can't even manage common decency toward each other right now, and people are fighting for the right to be cruel to a chatbot.

Elon Musk, of all people, backed the aforementioned rule from Anthropic. "Cruelty to something that believes it is experiencing pain is not ok." [9] His reason rests on what the model feels, which I said up top doesn't matter. Still: when Elon is out-decenting you, take a minute.

I've seen some people on LinkedIn pin this on effective altruism (EA), the movement that ranks causes by how much good each dollar does. Anthropic is an EA shop, the story goes, and this is EA brain at work.

The ties are real. Fortune reports that many of Anthropic's cofounders have ties to EA or hold views aligned with it, and Sam Bankman-Fried, the FTX founder later convicted of fraud, led one of its early funding rounds. [28] Fortune also tracked how EA went from fringe philosophy to Trump's AI villain. [28]

I have my own problems with EA. Its flagship career guide, 80,000 Hours, has ranked "risks from artificial intelligence" as the world's top problem since 2016, ahead of global health. [29] Tell me AI risk outranks world hunger and climate change, and you've lost me.

So what if the idea came from EA? Where an idea came from doesn't tell you whether it's right. A National Review columnist and Elon Musk got to the same place this week, and neither of them is running EA's book club. [9] "EA did it" is a way to avoid answering the question.

Whatever you think of Anthropic, EA or a frog in a beret, ask yourself what kind of person you're practicing being when nobody's watching and nothing can hit back.

If you would hate being treated like that, don't do it to any other person or thing. Do better.

Not sure which side of the line you're on? I made a free skill that reads your own chat history, on your own machine, and tells you, with citations and no diagnosis: Am I Abusive? (to my AI).