Three summers ago, Sam Altman, the C.E.O. of OpenAI, posted on X a fifteen-second video of dozens of people—mostly young, veiled in face masks, and wearing super-hip, white-soled sneakers—waiting in a long, snaky line in the traffic-light-lit but otherwise dusky dark of an unnamed city that appears to have been Nagoya, Japan. “Crazy lines around the world,” Altman wrote. “One person getting verified every 8 seconds now.” In the video, the camera pans to the front of the line, where a woman is kneeling on the sidewalk as if in prayer, staring into a gleaming silver orb about the size of a volleyball that has been propped up on a stand about the height of an end table. It looks like the eyeball of a giant robot Cyclops. She stares into its aperture, a void. You may wonder why. Here’s why. She is bowing before an artificially intelligent machine, seeking a certification that she is human.
That eyeball, which is known as the Orb, was developed by Tools for Humanity, a San Francisco- and Munich-based tech company that Altman founded in 2019 with a twenty-five-year-old German physicist named Alex Blania, who is now its C.E.O. At the time, Altman had just stepped down as president of Y Combinator to head OpenAI; Blania had dropped out of graduate school at Caltech. (Last year, Blania and Altman co-founded another startup, Merge Labs, funded by OpenAI, which, like Elon Musk’s company Neuralink, aims to hook your brain up to a computer. I guess this is another tool for humanity.) If you get scanned by the Orb, which you can now do at sites on four continents, you get a credential called a Proof of Humanity—provided, that is, that you are not a robot.
Are you a robot? If you’re not, you’re going to be pretty sure you’re not. Still, online at least, you likely find it difficult to tell who else is and is not. Businesses find this difficult, too, and it’s getting harder all the time. If you were forced to guess, the odds would tell you to guess robot, because at this point there are more bots online than people. To deal with this problem, businesses retain bot-detection and bot-management firms, a specialty within the field of cybersecurity. Their task used to be to block bots, but, with the rise of agentic A.I., that has changed. Now the problem is how to let in the good bots (like the avatar you send to the Gap to buy you a new pair of pants) and keep out the bad bots (like the thief trying to scrape the data of every Gap customer). The rest of us have to be our own bot detectors. It’s been suggested to me that if you’re on the phone or in an online chat with a customer-service “person” and you aren’t sure whether you’re talking to a human or a chatbot, one good idea is to ask, “What color shoes are you wearing today?” Because chatbots don’t have feet. (Yet.) I once asked a customer-service representative calling herself Crystal S. if she could prove to me that she was human, and she told me that she lived in Texas and had three daughters and two French bulldogs named Benny and Smoltz. I forgot to ask her about her shoes.
This is not a drill. But it is a hassle, something between an inconvenience and a nightmare, depending on your degree of enthusiasm for the shiny, glimmering how-we-live-now Orb of it all. It is also a very long game of cat and mouse. We used to be the cats. Now we’re the mice.
A certificate verifying your humanity is in certain ways a newer version of some ancient technologies of identification used not for distinguishing between humans and machines but for sorting one group of people from another. Those technologies include tattooing, cutting, and branding—methods of identification that go back thousands of years. In Revelation (13:16-17), the Beast requires everyone “both small and great, rich and poor, free and slave, to receive a mark on their right hand or on their foreheads.” The Greeks branded their slaves; the Romans tattooed their gladiators. In the twentieth century, the Nazis required Jews to wear badges, and then they tattooed them, “including newborn babies,” as Primo Levi once wrote. These techniques are still used for livestock, and for gang members.
Claiming to be a particular person has always been something of a performance, down to how you walk and talk and what clothes you wear. (Etymologically, the word “person” means “an actor’s mask”; that meaning was later expanded to describe “a character in a play” and, finally, simply, “a human.”) Imposture is a performance, too, as every con artist knows. A trickster named Arnaud du Tilh passed himself off as a peasant named Martin Guerre in sixteenth-century France by looking a fair bit like him and having a good memory for details. It was said in early modern England that you could pass yourself off as the king if only you could stop yourself from farting.
The more that populations grew, and the more that people moved around, the easier it was to pretend to be someone else. Hence, identity documents, beginning with travel papers. Medieval merchants and soldiers were often required to carry laissez-passers. “Where passports do not exist, nothing is easier than to change one’s name,” Tocqueville later observed. Proof of identity, and especially proof of citizenship, came to require written verification in the nineteenth century, when documents like birth certificates and passports became commonplace; they frequently required seals and stamps and signatures by representatives of the church and later of the state. Like money, these documents could also be faked. Signatures could be forged; photographs could be altered. Fingerprints proved more reliable.
By the end of the nineteenth century, the rise of the social-welfare state meant that Britain, Germany, and France began keeping files on individual citizens for purposes of the provision of assistance. Personal identification numbers became common in the nineteen-thirties and forties. (Meanwhile, totalitarian regimes that assign people numbers instead of allowing them names have been a hallmark of dystopian fiction at least since Yevgeny Zamyatin’s 1924 novel, “We.”) Nazi Germany pursued a policy of Erfassung, total documentation of the population. “We are recording the individual characteristics of every single member of the nation onto a little card,” the head of an I.B.M. subsidiary in Germany explained in 1934. The United States introduced the nine-digit Social Security number in 1936, partly to solve what was called “the Fred Smith problem” (the fact that there are a lot of people with the same name). Critics charged F.D.R.’s federal government with despotism, and the head of the Republican Party warned, “If the Roosevelt administration is returned to power, we shall see two groups of citizens in this nation: those who are numbered and those who are not numbered.” But plenty of Americans carried their Social Security cards with pride, and even displayed their numbers on jewelry and as tattoos, believing the number to be, as the historian Sarah Igo has argued, “a symbol of a new kind of economic citizenship.”
Sweden assigned a personnummer to all citizens beginning in 1947; Icelanders were assigned an eight-digit number in the early nineteen-fifties, when each person in the country was given a punch card in the National Register. By that decade, governments were storing personal information on computers for purposes like keeping track of tax payments and providing benefits to veterans. (In the nineteen-sixties, a planned National Data Center in the U.S., merging all this information together, was scuppered because of privacy concerns.) Starting in the nineteen-eighties, forms of identification as proof of citizenship became more rigidly required. Today, most nation-states require citizens to register with a national database and have a national identity document. The technologies supporting them include R.F.I.D. chips and facial-recognition software. Malaysia introduced biometric passports in 1998. Mexico began issuing national I.D. cards incorporating iris-scan data in 2011. In 2010, the United Kingdom abandoned a plan to introduce national I.D. cards that would have included biometric data. India’s Aadhaar system, created in 2009, uses iris scans and fingerprints, and assigns a unique twelve-digit number to all residents. As of 2026, when anyone from outside the E.U. enters nearly any E.U. country, no nice passport person says bienvenue or willkommen; instead, a machine scans your passport, your face, and your fingertips, to verify that you are you.
Online, though, you can be anyone. The opening of the internet to the public in the nineteen-nineties led to new forms of identification—including usernames and passwords and two-factor verification and Touch ID—and new kinds of fraud and, inevitably, new forms of imposture, as in: pretending to be made of flesh and bone when, really, you’re some blocky lines of machine code.
Science-fiction writers saw this coming long before it came. An Isaac Asimov story from 1946, set in the future, tells of an ambitious district attorney named Stephen Byerley who is suspected by political opponents of being a robot. (Decades later, some people said the same thing about Al Gore.) “He has never been seen to eat or drink,” a rival says. “Never!” That is insufficient proof, but in Asimov’s future world there’s another kind of test available: harming a human is a violation of one of the Three Laws of Robotics, so, if Byerley can be shown to be unable to harm a human, he must be a robot. At a campaign rally, a man approaches Byerley and says, “Hit me! You say you’re not a robot. Prove it.” Byerley smacks the schmuck in the face and wins the election. Nobody had considered the possibility that it was all a setup planned not by Byerley’s enemies but by Byerley himself, and that the schmuck was . . . a robot. In a later story, Byerley’s political career takes off and by 2044 he is elected essentially President of everything, the humans still not having worked out that their leader is probably a robot. Fortunately, this no longer much matters since, as a robopsychologist reflects, “by that time it was the Machines that were running the world anyway.”
Cold War science fiction was consumed with the idea of robots pretending to be humans, the way Communists pretended to be Americans, everyone a suspected subversive. In a 1953 story by Philip K. Dick called “Impostor,” a defense-project official named Spence Olham is killed and replaced by a robot, who has a “U-bomb” inside him, but the poor robot thinks he’s the real Olham:
(The proof, or disproof, comes when he explodes.)
Most of these stories involving tests are variations on the seventeenth-century philosophical meditations of René (“I think therefore I am”) Descartes. Seeking certainty about his own existence, Descartes concluded that the very act of thinking proved that he existed as a thinking being. As Anthropic’s chatbot, Claude, will tell you, this proof is no longer sufficient and hasn’t been since Alan Turing proposed the Turing test, in 1950. That led to new tests. In 1968, in Philip K. Dick’s novel “Do Androids Dream of Electric Sheep?”—later adapted into the film “Blade Runner”—a character named Rick Deckard (an homage to Descartes) administers the Voigt-Kampff Empathy Test to suspected androids. The test involves measuring a subject’s reactions to scenarios of cruelty to animals: humans react instantly, but it takes androids a fraction of a second longer to register that they are supposed to find these stories upsetting. The test works, the narrator muses, because “empathy, evidently, existed only within the human community, whereas intelligence to some degree could be found throughout every phylum and order including the arachnids.”
It appears that a rosier view of humanity prevailed in the nineteen-sixties than prevails on Sam Altman’s planet Earth. These days, to find out if you are or are not a robot, no one is going to ask you if a photograph of beagles with their heads trapped in a vise being forced to smoke cigarettes disturbs you. They may, however, ask you to look into an Orb.
Online _“_I’m not a robot” tests date to the earliest forms of a protocol that came to be called a CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart), first imagined in 1996 by an Israeli cryptographer named Moni Naor and developed at Carnegie Mellon by Luis von Ahn, who started work on it when he was only a graduate student. (Von Ahn also came up with the name.) “How do we prevent computer programs from using systems designed for human use only?” Carnegie Mellon’s CAPTCHA Project wanted to know. By hand, this task is impossible, given the scale of the problem. “A better solution enlists computers to fight computers: write a program that can recognize, through a brief interaction over the web, whether a user is human or not.”
CAPTCHAs were developed to deter and detect spam, worms, and bots, and, later, to defend against disinformation campaigns, election interference, and acts of cyberterrorism. Most early CAPTCHA tests involved a computer asking you to prove that you are not a robot by deciphering a patch of blurry, crinkled, or otherwise difficult-to-decipher text. These tests soon began to pay a surprising dividend: they allowed Google to crowdsource the optical-character-recognition work it needed to scan the world’s collection of printed books (without, it must be said, the permission of publishers or authors). In 2007, von Ahn, who had become a professor, launched reCAPTCHA, with the slogan “STOP SPAM. READ BOOKS.” Google purchased reCAPTCHA in 2009.
With reCAPTCHA, every time a human resolved a blurry squiggle or a broken word or phrase, that human was advancing the Google Books Project. The same applied to the maddening visual tests—find the traffic lights, the crosswalks, the bicycles, and teach computers to see better. Von Ahn, who left academia and went on to found Duolingo, has expressed regret about the puzzle phase of proof of humanity, wondering whether he had created an apparatus for siphoning off human attention on a vast scale, a few seconds at a time. But, looked at another way, all that puzzle solving wasn’t a waste of our time if you consider how effective we were at teaching machines how to replace us.
The puzzles, in any event, became obsolete. The first chink in the armor came when (human) crooks started assembling CAPTCHA farms. They’d hire thousands of (real) people to staff their operation in, say, the Philippines, and, if a bot hit a CAPTCHA test, it’d call a human to crack it and then: open sesame.
It was to defeat CAPTCHA farmers and other no-goodniks that, in 2012, Tamer Hassan and three friends started a bot-detection company that’s now called HUMAN, while hanging out in the back of Singularity&Co., a science-fiction bookstore in Brooklyn. (Singularity was less a bookstore than an archival philanthropy: with money from a Kickstarter campaign called Save the SciFi, they’d buy up rare, out-of-print twentieth-century pulp science fiction, scan it, and then sell the original books, by the pound.) Hassan was newly back from a tour in Iraq with the Air Force, flying helicopters and C-130s. He started HUMAN, he told me, because he and his co-founders wanted “to disrupt the economics of cybercrime.” Cybercrime comes in all shapes and sizes, but it does not require a great deal of capital to do it. “You get teen-age hackers in their pajamas,” Hassan says. “You get college kids trying to get extra money for drinks on the weekend and you get organized cybercrime and you get terrorist groups.” Law enforcement hadn’t worked especially well. And CAPTCHA was failing, too.
What began to replace asking humans to complete puzzles was asking computers to monitor internet traffic more closely. Beginning in 2014, Google launched a series of tests without puzzles (a class of tests that came to be known as “invisible reCAPTCHA”). You see a box that you can click to certify “I’m not a robot”—as if anyone would take your word for it!—but the box is essentially a ruse, because, while you’re dragging your cursor over to click on it, the program is already determining whether or not you’re a robot by evaluating your past and present online behavior: where you click and have clicked, how long you , what cookies you’ve collected, your I.P. address, how you move your mouse, how you scroll. “In the early days, you could spot something that wasn’t real with basic algorithms,” Hassan says. “If somebody moves too fast through the pages, you can say that is nonhuman.” Humans are messy; bots are tidy.
This worked well for a long time because bots used to be stupider, and they also used to be more expensive. “Maybe ten years ago, it was pricey to build a really good bot, a really efficient bot,” Benjamin Barrier told me. Barrier co-founded an international bot-detection company called DataDome in 2015, in Paris. (The company now also has offices in New York and Singapore.) Bots got smarter and cheaper, and then you could buy what came to be called “packaged bots,” something akin to a full-service bot provider. These days, Barrier says, “you spend a few hundred bucks a month and you can rent out a perfect bot,” or, more likely, an army of bots. It’s the difference between one rat in your kitchen and a rat infestation in your basement.
Packaged bots are the precursor to today’s “A.I. agents,” which is just a name for bots that aren’t doing things that are illegal or socially or commercially undesirable. Some people keep rats as pets! So now the problem, for companies like HUMAN and DataDome, is figuring out a bot’s intent—is this one a pest or a pet?—which is quite a bit more difficult than just keeping them out. That requires analyzing behavior, assembling a library of all known bad bots, and using A.I. models and cryptography protocols to tell the bad bots from the good bots. “That time when bots were bad and humans were good, that was a much easier time,” Barrier says with a sigh.
DataDome promises to help clients “regain control of your traffic.” But the metaphors in the bot-management industry lean less toward air-traffic control than toward national defense. HUMAN has both a military and a public-service sensibility. Stu Solomon, HUMAN’s current C.E.O., is, like Hassan, an Air Force Academy graduate; he is also a brigadier general in the Air National Guard. Solomon and Hassan’s interest in cybersecurity comes from their commitment to the prevention of crime, counterfeiting, fraud, and terrorism. (Before joining HUMAN, Solomon was a cybersecurity executive at Bank of America.) If this is a war, it’s over, and the humans lost. It’s no longer possible to defend the internet against bots. Solomon says, “Now we need to think about this as coexistence.”
Solomon thinks the fact that there are now more bots than humans online was always inevitable. Hassan agrees, and he also thinks it would be best if, at this point, humans simply cede the internet to the robots—the good ones. “The internet was made for machines to talk to each other,” he says. “It’s a system of protocols for computers to interact and message each other. In the end, that’s what it will be. It will be an internet of A.I. agents, operating on our behalf. Probably in protocols and with languages that we don’t understand anymore.”
He’s O.K. with that. We still have the real world all to ourselves. Don’t we?
The Orb is not a bot-detection device. It’s a human-capturing device. Sam Altman and Alex Blania’s company Tools for Humanity has taken an entirely different approach to cybersecurity than companies like HUMAN and DataDome have. Tools for Humanity is interested in securing onetime, portable proof of your humanity and charging other companies for the use of that evidence. The Orb is different in other ways, too. Rather than having origins in crime prevention or cybersecurity or national defense, the Orb has origins in cryptocurrency.
The first tool for humanity that was forged by Tools for Humanity, in 2019, was a cryptocurrency called Worldcoin. Actually, that was the purpose for which the company was founded, because the whole scheme started out as what Blania refers to as “the Bitcoin Project,” a universal-basic-income proposition. The original idea had been to hand out Worldcoin to every human on the planet as a hedge against the revolution otherwise inherent in the coming A.I.-driven jobpocalypse. Start handing out money, and you need to verify everyone so that no one double dips, and so that robots don’t dig around in the till with their grubby little robot paws. In a public announcement in 2021, the company described Worldcoin as a “new, collectively owned global currency that will be distributed fairly to as many people as possible.” Get scanned, get verified, get paid. Revolution averted.
Both the Orb and the story that Tools for Humanity tells about its purpose have changed since then. An early version of the Orb, which looked like a miniature Death Star, spat out physical coins, like a slot machine, presumably to make the point about the quid pro quo to people who might not quite comprehend blockchain. Prior to the product’s launch, the field testing spanned thirty-four countries—many of them developing nations—and included such cities as Buenos Aires, Lisbon, Bangalore, and Nairobi. Worldcoin also gathered training data for its A.I.-driven Orb by sending Orbs to locations in Sudan, Chile, and Ghana. The MIT Technology Review reported that villagers in Gunungguruh, Indonesia, lined up for hours to have their irises scanned, having been offered “everything from free cash (often local currency as well as Worldcoin tokens) to AirPods to promises of future wealth.” (A company spokesperson responded to the Review, saying that the Orb operators were “independent contractors, not Worldcoin employees,” and that the giveaways were “isolated efforts” on the part of these workers.) At the time, twenty-five tokens was a typical payment; today, that’s worth about eight dollars. The project’s ambitions swelled like a tick on a dog’s belly. When the token publicly launched, in July of 2023—a time when OpenAI was doing a lot of posturing about saving democracy, currying favor with the we-are-saving-democracy Biden Administration, and only months before the board of OpenAI tried to oust Altman on the grounds that he was reckless and untrustworthy—Altman and Blania said, in a statement, “If successful, we believe Worldcoin could drastically increase economic opportunity, scale a reliable solution for distinguishing humans from AI online while preserving privacy, enable global democratic processes and eventually show a potential path to AI-funded UBI (universal basic income).” A rollout that summer focussed on higher-income nations in Asia, especially Japan and South Korea, where people who queued at pop-up Orb stations, like the one Altman tweeted about, were paid in Worldcoin. Then the tick burst. Later that year, talk of universal basic income disappeared from Worldcoin’s website. Soon, Tools for Humanity dropped the “coin” from the Worldcoin name and, for the most part, stopped mentioning crypto.
The latest version of the Orb is meant to look less sinister, more like something made by Apple or, as Tiago Sada, Tools for Humanity’s chief product officer, told me, “like something out of Pixar.” That cute little gooseneck desk lamp, say, or EVE, the gleaming-white legless robot from “WALL-E.” That scarcely disguises its darker implications for human freedom.
One of the thornier political, legal, and moral concerns raised by companies like OpenAI and Tools for Humanity is the emergence of what I think of as the “artificial state”: for-profit, international corporations assuming powers once reserved for nation-states, without any democratic mandate. With the Orb, Tools for Humanity proposes to take over the historically governmental function of identification.
World wants everyone in the world to get scanned. When you get scanned, you verify your WorldID, and with the World App you buy gift cards and cryptocurrency. This hasn’t gone over very well. A sizable number of nation-states, including Argentina, Brazil, Colombia, Germany, Indonesia, Kenya, the Philippines, Portugal, South Korea, Spain, and Thailand, have either investigated or banned the Orb or fined Tools for Humanity, citing privacy or consumer-protection violations, ethical misgivings, or regulatory concerns.
Nevertheless, the Orb may be coming soon to a coffee shop, sidewalk, or shopping mall near you. “We have about forty million people on the network with some level of verification and about half of those, eighteen million, at the highest level of verification—the Orb verification,” Sada told me. “But right now we are really focussed on the U.S.” During the Biden Administration, the S.E.C. wasn’t inclined to look favorably on people being paid in crypto for having their eyeballs scanned. That changed when Donald Trump returned to office, and since the spring of 2025 it has been possible to secure a Proof of Humanity from an Orb within the land of the free and the home of the brave.
During the past year’s P.R. push, people at Tools for Humanity and World have talked less about universal basic income, or democracy, and more about protecting humans from a world overrun by artificial intelligence, especially in the form of bots, scams, malware, deepfakes, and other assorted automated nuisances that are swarming all over the internet. Sada, who is from Mexico City, started building robots when he was thirteen. He met Altman when he was briefly an undergraduate at Worcester Polytechnic Institute: Y Combinator funded an app Sada invented, called Flinto, that is akin to Venmo but for Latin America. He joined Tools for Humanity in 2021, when he was twenty-four. Sada told me that Altman and Blania started Tools for Humanity because they were concerned that, one day, “trust on the internet would start to get eroded.”
Essentially, World, which has about two hundred and fifty million dollars in venture-capital funding, proposes an alternative internet—or at least an exclusive space within the existing internet, something like a first-class airport lounge—for humans only. Its motto is “Let’s make the internet more human.” (This, mind you, is coming from the people who brought you the bots in the first place.) But, to get into that lounge, you need to get scanned and verified by the Orb. This strategy has achieved some success. WorldID is now integrated, as a human-verification mechanism, into apps that include Zoom, Tinder, Shopify, and DocuSign. Verifying more humans is Blania’s whole reason for living, he told Tools for Humanity’s employees last year: “That’s all I care about every day and all you should care about every day, and nothing else should matter.” Since then, given the growing number of countries questioning the Orb, the company has appeared to be struggling. I talked to Sada at the end of May. In early June, the day OpenAI filed paperwork for going public, Tools for Humanity, valued at $2.5 billion, announced that it was laying off staff because of a shifting “company strategy.”
None of this has quieted the hype, or the revisionist history. This spring, at a World event in San Francisco celebrating the launch of a new version of WorldID (“We think it’s a really special moment in human history,” one company representative said), Altman made no mention of World’s original plan—handing out money to people put out of work by A.I. Instead, he insisted that he and Blania had started on this path years ago because they knew advanced artificial intelligence was coming, and they wanted to preserve humanity in a world “where there’s going to be more stuff generated by A.I. than by humans.” A world overrun with bots is a wonderful world, in Altman’s view—so long as they are doing our bidding. There’s just this one hitch, he admitted: “I’m sure many of you have started to have an experience more and more where you’re, like, ‘Am I interacting with an A.I. or a person?’ ”
It’s not a speech that schoolchildren will one day memorize, like when Abraham Lincoln resolved “that this nation, under God, shall have a new birth of freedom—and that government of the people, by the people, for the people, shall not perish from the earth.” No, this wasn’t Lincoln at Gettysburg. This was Altman in the Valley. “We really wanted to make sure that the world continued to be about people, for people, that people were still the primary thing that we all care about,” Altman said. Are you reassured? ♦