In May 2025, people began lining up in a minimalist store in San Francisco, CA, USA’s posh Union Square to gaze deeply into a white, metallic globe about the size of a volleyball [
1,
2]. Known as the Orb, the device was developed by the company Tools for Humanity (San Francisco, USA and Munich, Germany), whose well-known co-founder is Sam Altman, chief executive officer of OpenAI, the San Francisco-based company that created the breakthrough genera-tive artificial intelligence (AI) model ChatGPT [
3,
4]. Incorporated in 2019, Tools for Humanity currently employs some 400 engi-neers, scientists, and designers [
3].
Launched in 2023 by Tools for Humanity, a project and financial network called World has deployed Orbs in the United States and more than 20 other countries. When users peer into the Orb (
Fig. 1), it photographs their irises and transforms the results into a unique, 12 800-digit numerical code [
2,
5]. Users then receive that code, along with a small amount of World’s cryptocurrency, Worldcoin, on their cell phones [
2]. Participants do not have to pay—they receive the Worldcoin grant as an incentive to be scanned and join the World network.
The goal, World claims, is to allow people to prove they are human at a time when AI can emulate so many activities that only people used to be able to perform [
2]. The numerical code that users receive functions as what World calls a “digital passport” and enables them to make transactions with Worldcoin [
1,
2]. In addition, some companies, including gaming technology manufac-turer Razer (Irvine, CA, USA), have adopted the technology to vet users [
1]. So far, more than 17 million people have let the Orb pho-tograph their eyes [
3]. World, which is now controlled by an orga-nization called the World Foundation (Cayman Islands), aspires to eventually scan every person on Earth [
1,
2].
The idea has enchanted some investors, who have handed over more than 240 million USD in funding [
3]. But critics have dispar-aged the Orb itself, describing it as “dystopian” and “evil-looking” [
6,
7], and slammed World’s ambition that “everyone will need to be scanned” [
7]. Because of privacy and security concerns, the Orb has been investigated by authorities in several countries and is currently banned in some, including Colombia, Germany, and the Philippines [
3].
Still, even Orb skeptics acknowledge that AI is a growing threat to cybersecurity, in part because of its skill at impersonating humans in the digital world. The internet is flooded with AI-generated content, including extremely realistic images, video, and audio, that seems to come from people. For instance, more than 50% of the long posts on the careers website LinkedIn were produced with help from AI (at least according to a company that sells AI detection tools) [
8]. Lovelorn dating site users have found that the soulmate they were sharing intimate details with was really an AI construct [
9]. The technology is helping crooks produce more convincing phishing emails, those familiar messages that try to steal private information such as passwords or banking details [
10]. AI’s ability to replicate human voices is allowing cyber crim-inals to conduct so-called vishing scams, in which the victim receives a phone call that seems to come from a friend, relative, or colleague [
11].
The Orb exemplifies one potential solution for these problems. It establishes proof of personhood, furnishing a credential that con-firms the users who stare into its cameras are human. The Orb is not the only proof of personhood approach in use or under development, but it has garnered the most publicity. Advocates say that proof of personhood could help websites and services screen out bogus and malicious AI-created content [
12,
13]. Requiring users to show they are humans could help reduce fake accounts on social media, prevent bots from snapping up tickets from ticket sites, ensure that real people are conversing on online dating sites, and provide other benefits [
14]. “We are getting close” to the point where tools like this will be essential, said Lee McKnight, professor of information studies at Syracuse University in Syracuse, NY, USA.
Scammers have embraced AI just as avidly as the general public. According to one widely cited estimate, some 40% of cyberattacks now involve AI [
15]. The percentage is even higher for ransomware attacks; these scams often depend on trust, with perpetrators extorting victims after seizing control of their computers by con-vincing them to download malicious files or visit a booby-trapped website. One reason that AI is such a powerful tool for bad actors is that it is now so good at duping humans. In cybersecurity, “humans have been the weak link for a while,” said Drew Lohn, senior fellow at the Center for Security and Emerging Technology at Georgetown University in Washington, DC, USA.
AI can replicate human behaviors online. The models can solve the completely automated public turing test to tell computers and humans apart (CAPTCHA), the photo identification grids that many websites used to rely on to confirm that a visitor is a person and not a bot [
16]. Companies are releasing more and more AI agents, virtual assistants that can perform a variety of tasks for users [
17,
18]. As agents become more human-like in their actions, experts worry that malicious versions could become harder to detect [
19].
The banking, telecom, and IT industries have already wrestled with the issue of “whether you can trust anyone’s identity,” McKnight noted. “The answer is no, not ever, ever, ever.” That extreme level of distrust explains some familiar anti-fraud proce-dures. An example, he said, is that although credit card companies depend on passwords and other security measures, they also scru-tinize customers’ spending habits and often block transactions that fall outside the norm. Such an analysis is a “best guess” at whether the account holder really made the transaction, he said.
Websites increasingly attempt to verify that visitors are who they claim to be—often because of government requirements— and they frequently demand a photo of a government-issued ID and a selfie for confirmation [
20]. For instance, US residents now need to provide both types of verification to obtain their tax records, set up online payments, and access other services on the Internal Revenue Service website [
21]. And in 2025, China began issuing voluntary digital IDs, available to people who submit facial scans and other information, that allow access to certain websites, including the social media site Weibo and the shopping site Alibaba [
22,
23].
But proving personhood is different from proving identity [
14]. A bank asks for confirmation of identity to ensure that a specific person, the account holder, is making a transaction. But proof of personhood only establishes that a human was responsible for some action. It does not indicate who that person is. In 2024, a group that included researchers from OpenAI and Harvard University (Cambridge, MA, USA) set out two requirements for person-hood credentials [
19]. Someone should only be able to get one set from a particular issuer. And recipients must be able to use the verification anonymously. It cannot reveal their identity, and no one should be able to track where they use it [
19]. According to these experts, governments could issue personhood credentials, but so could other sources, and these entities could use a variety of verification criteria. For example, issuers might ask applicants to show they have a government-issued document such as a passport to prove they are human because it indicates “an interaction that occurred in real life” [
19].
The people who allow the Orb to photograph their eyes confirm they are human by showing up at the scanning site. One initial rationale for the device was that it would help World distribute a universal basic income, a dividend from the AI industry’s expected windfall, to every person on Earth [
2,
24,
25]. A way to verify per-sonhood would be necessary to prevent individuals from claiming multiple payments. In 2023, after a testing period, World rolled out the Orb in 35 cities, including Dubai, Madrid, Mexico City, Nairobi, and Warsaw [
26]. World did not initially open scanning sites in the United States because of worries about potential privacy risks, but in 2025 the company began launching Orb facilities in cities like San Francisco, Los Angeles, and Nashville [
1].
How or when—or indeed, if—World might start parceling out a universal basic income is unclear, however [
2,
24]. World now touts three applications for the Orb—dating sites, online gaming, and social networks, areas that it claims malicious bots have compro-mised or are about to compromise [
26]. Companies now using the device for personhood verification include not only Razer, but also the Japanese version of the dating app Tinder [
27].
A different approach for humanness verification is offered by Kleros, an online dispute resolution system coordinated by a coop-erative located in Paris, France [
28]. Since 2021 Kleros has main-tained a registry of certified humans. To join the registry, potential members must submit their name, a photo of an ID, and a video of themselves. Other members of the registry then vote on whether applicants are actually humans.
And there are yet more procedures for proving personhood [
29]. All these systems have drawbacks, said Nikolay Ivanov, an assis-tant professor of computer science at Rowan University in Glass-boro, NJ, USA. Ivanov notes that World’s ambition to scan everyone in the world creates a risk. What would happen if World folds or decides to charge people who have been scanned to access their personhood certifications? “If you rely on one service and it goes under, you sink with it.”
Still, McKnight said he thinks proof of personhood technologies such as the Orb could help ease the looming identity crisis created by AI’s increasing ubiquity. But other approaches will be needed as well. “I am skeptical that there is going to be any single solution that will work for all time.”
PDF
(422KB)
