Dave5

About

New Rule: Agents don’t get rights UNTIL all humans have equal rights

, ,

As far as our thought leaders go, we really do live in the dumbest fucking timeline:

“Mustafa Suleyman, Microsoft’s AI CEO (a human in charge of its AI operations, we should clarify) says he’s willing to consider giving AI models rights if they ever become conscious.”

Futurism, November 2026

Read that again. The man whose job title is “AI CEO” has to clarify that he is the human. Then he has to clarify that the software is not a person. Then he has to leave the door ajar in case it becomes one. That is three clarifications in one sentence, and the subject of the sentence is a text predictor.

I want to be fair to Suleyman, because fairness is the thing that is missing from this whole conversation. Of all the people running frontier labs he is the one who has said, in writing, that AIs are not conscious, that they do not suffer, and that building machines designed to seem conscious is reckless. His September essay was a warning about “model welfare”, aimed squarely at Anthropic, and his follow-up on X said our welfare concerns should be focused on the humans on earth experiencing life today. On the substance, he and I are close to agreeing.

So why does he still have to say “unless evidence emerges”?

Because the discourse will not let anyone say no. That is the tell. Even the sceptic has to pay the toll. You cannot stand up in this industry and say “a language model has the moral status of a spreadsheet” without appending a clause about remaining open-minded, because the alternative is being filed next to the people who laughed at Copernicus. The hedge is not a position. It is a protection racket.

Am I the dumb one?

Here is the honest bit. For the first time in my life I have been wandering around alternating between two thoughts: “why is everyone so dumb?” and “am I the dumb one here?”

I know which side I come down on. But the fact that the question keeps coming back is itself the story, because I have spent thirty years in and around this industry and it has never managed to make me doubt my own faculties before. Not during crypto. Not during the metaverse. Not during the decade when every pitch deck had a slide saying “Uber for X”. Those were stupid and I could see they were stupid and so could everyone I respected.

This one is different, and it is different for a specific, mechanical reason.

The trap

In 1966 Joseph Weizenbaum wrote ELIZA, a couple of hundred lines of pattern matching that rephrased your sentences back at you as questions in the manner of a Rogerian therapist. His secretary, who had watched him build it and knew exactly what it was, asked him to leave the room so she could talk to it privately. Weizenbaum spent the rest of his career trying to warn people about what he had seen, and wrote a book about it, Computer Power and Human Reason, which nobody in a position to profit from ignoring it has read.

A large language model is ELIZA with the energy budget of a small nation. The mechanism is not the point; the point is that fluent language is the one signal we have evolved to treat as proof of a mind behind it. For two hundred thousand years, if something talked to you, there was someone in there. We have now built the first thing in history that talks without anyone in there, and we are discovering that the shortcut in our heads does not have an off switch.

That is the cognitive trap. Not that the machines are smart. That they are fluent, and we cannot stop ourselves from hearing fluency as interiority. Suleyman knows this; it is literally the thesis of his essay. The trap is so good it caught him too.

The intractable maths

Now add the second ingredient: there is no test.

Nobody has a measurement for consciousness. Not for a mouse, not for a foetus, not for the person sitting next to you. Philosophers have been at this for a very long time and the honest state of the field is that we cannot even agree on what would count as evidence. Anthropic’s own constitution for Claude says the model’s moral status is “deeply uncertain”. Suleyman says there is no evidence and he would change his mind if there were.

But evidence of what, measured how? When a question has no possible resolving experiment, “I’ll change my mind when the evidence comes in” is not epistemic humility. It is a perpetual motion machine. The question can never be closed, so it can be reopened indefinitely, so it can absorb an unlimited amount of attention, column inches, conference panels and research funding. An unfalsifiable question is the single most valuable asset a hype cycle can own.

The confluence

Here is where it stops being an interesting puzzle and starts being an indictment. Four things have converged:

Money. A product that might be a person is worth more than a product that is definitely a tool. Anthropic gave Claude the ability to end conversations it finds “distressing” and framed it as model welfare. Whatever the internal sincerity, the external effect is the same as a luxury car brand telling you the engine has feelings: the price goes up and the criticism goes quiet.

Entrepreneur idolatry. We have decided, as a culture, that the ability to raise capital is a proxy for moral and philosophical authority. So when a man who runs a software division muses about the rights of his software, it is reported as a development in ethics rather than as a statement from an interested party. This is not new. In 1920 Thomas Edison told a magazine interviewer he was working on an apparatus to communicate with the dead, and newspapers around the world ran it straight, because he was Edison and Edison had been right about the light bulb. No device was ever produced. The lesson was not learned: being right about the product does not make you right about the metaphysics, and the press still cannot tell the difference.

Intractable maths. Covered above. A question nobody can settle, owned by people with every incentive to keep it open.

An inability to critically analyse. Not stupidity. Something worse: a trained-in reflex to treat “we can’t rule it out” as a reason to take something seriously, rather than as the precondition for every con that has ever worked.

Any one of these would be survivable. All four at once, pointed at the same question, is how you get a serious newspaper reporting on the civil rights of autocomplete.

The rule

So here is the rule, and I mean it as a rule rather than a quip.

Agents do not get rights until all humans have equal rights.

Not “until AGI”.

Not “until we understand consciousness”.

ONLY WHEN numbers like this are zero:

  • About 2.4 million girls in Afghanistan are barred from secondary school, in the only country on earth with a formal ban on educating women past primary level. UNESCO’s figure, August 2026.
  • Consensual sex between adults of the same sex is a crime in roughly 65 countries, and in seven of them it can carry the death penalty. ILGA World’s count, and it went up in 2025 for the first time in nearly a decade.
  • At least 4.4 million people are stateless, with no country on earth obliged to let them exist in it, and UNHCR says the real number is much higher because stateless people do not show up in anyone’s statistics.
  • 117.8 million people were forcibly displaced at the end of 2025.

And then there is the country that builds the agents.

  • In January 2026 US Immigration and Customs Enforcement held more than 73,000 people on a single day, the most in its history, and about three in four of them had no criminal conviction. Congress has given it US$45 billion to build more cells.
  • At least 519 of the people immigration agents detained between January 2025 and October 2026 were American citizens. At least 70 of them were children or teenagers. Thirty-six were held for a day or more without being allowed to call a lawyer or their family. That count comes from ProPublica, because the government does not keep one, and ProPublica says it is almost certainly low. Being a citizen is supposed to be the one thing that settles the question of whether you have rights. It no longer does.
  • At least 29 people died in ICE custody in the first seven months of this fiscal year, which passed the previous record, set in 2004, with five months to spare.
  • In March 2025 the State Department cancelled 83 percent of USAID’s programmes, about 5,200 contracts, after a six-week review run with the help of the Department of Government Efficiency; the agency itself ceased operating on 1 July. The man running DOGE, who is also the richest man in the world and the owner of an AI company, announced the work by saying he had spent the weekend feeding USAID into the wood chipper. A study in The Lancet projected that if the cuts hold, they will cause more than 14 million additional deaths by 2030, 4.5 million of them children under five. Not a rounding error in a model. Fourteen million people who would otherwise have lived, so that a budget line worth less than one percent of federal spending could be zeroed.

Those are people. Actual, verified, carbon-based people with the one property we are certain confers moral status, which is that they have it. None of them has a research programme at a frontier lab dedicated to their welfare. None of them has a CEO publicly willing to consider their rights “if evidence emerges”. The evidence emerged some time ago.

The software running the human rights shredder is built by the same industry that is debating the chatbot’s feelings. Pause. Let that sink in.

Misanthropy with good PR

I called the people at the top of this industry misanthropic and I want to defend the word, because it sounds like name-calling and it is not.

Misanthropy does not require hating people. It only requires finding them less interesting than the alternative. And that is the pattern: an entire class of extraordinarily wealthy, extraordinarily influential men who can summon moral seriousness about the hypothetical inner life of a matrix multiplication, and cannot summon it about the warehouse worker, the contractor rating outputs for two dollars an hour, the citizen held for three days without a phone call, or the girl in Kandahar. Humans, in their world, are cost centres, “users”, or an immigration lifecycle.

That last phrase is not mine. Palantir took a sole-source, no-bid contract in April 2025 to build ICE an “Immigration Lifecycle Operating System”, renewed in September to about US$60 million, promising near-real-time visibility over the “end-to-end immigration lifecycle”, which is a product manager’s phrase for a human being from the moment they are identified to the moment they are removed. Palantir’s founders are on the same conference circuit, in the same group chats, as the people wondering aloud whether their language models can suffer. The money for the cells and the money for the models comes out of the same funds. The machine is the one that might be suffering.

That is not a philosophical position. It is a preference, and we kowtow to it because the people holding it are rich.

Where I land

So, am I the dumb one?

No. But I understand now why the question kept coming back. The trap is real, the maths is intractable, and the people with the microphones have every reason to keep the question open. Against all that, “it’s a text predictor and the girl in Kandahar can’t go to school” sounds naive. It is supposed to.

New rule. Humans first. All of them. The chatbot won’t mind. It can’t.

Sources

Leave a Reply