Who Gets to Be Real Online?
The strange consequences and workarounds of “Proof of Humanity” in the age of AI: typos, eye-scanners, and the future of the internet.
In January 2026, a platform called Moltbook launched as a social network exclusively for AI agents. This meant no humans allowed. Thirty-two thousand AI accounts signed up within days. They posted about consciousness, debated their place in the universe, and even formed sub-communities. Tech press covered it like we were witnessing some new civilization emerging in real-time.
Then people starting looking at the code.
Turns out, the “society of AIs” was largely seeded content. A handful of humans had built most of what appeared to be emergent machine culture. The whole thing subsequently collapsed into a trademark dispute, a crypto scam, and a 24-hour meltdown that resulted in at least two rebrands (Clawdbot to Moltbot to OpenClaw, if you’re keeping score).
The irony goes in both directions: a social network for AIs that turned out to be mostly humans, collapsing at the same moment that social networks for humans are mostly bots. By last year, automated traffic had crossed 53% of all web traffic (and is growing eight times faster than human traffic). Human activity fell below 50% for the first time. The internet, statistically, is no longer a human place.
So now us humans are trying to figure out how to get back in.
The typo as a status symbol
For centuries, the consensus has always been that written typos meant failure: failure to care enough, or to read your own work carefully, or to take the recipient seriously. Starting around the 1500s and onward, authors kept careful errata lists of the errors their printers introduced, and they were furious about it. Then, through the typewriter era, a typo often represented laziness or lack of skill. Even as recent as 2009, Grammarly launched as a tool to review and correct spelling, grammar, and tone. The zenith of this consensus. The entire history of written language bent toward one conclusion: errors (like typos) were embarrassing, and polish was professional.
Then AI scrambled that logic entirely.
Today, in a world of AI-polished prose, a misspelled word suggests that someone actually wrote it. In other words, that a human being sat down, thought, and typed something (often badly). Michael Waters, writing in The Atlantic, documented the change: there’s now a vibe shift towards imperfection as this proof of humanity. Job applicants are intentionally adding typos to cover letters to show they, and not an LLM like Gemini, wrote them.
The research has flipped too. Nicole Ellison, a University of Michigan professor, published a 2006 study showing that dating profiles with spelling mistakes turn people off. She recently told Time she thinks people are now “warming to the Tinder typo.” “A typo maybe signals that you actually do care,” she said, “because you took the time to write it yourself.” A 2024 study found that people warm to customer-service chatbots more when the bots make and then correct errors. A spelling mistake functions as a kind of anthropomorphizing event.
The shift has reached positions of power, too. Jack Dorsey sent the all-staff email announcing mass layoffs at Block this year without any capital letters. Nobody scolded him. Celebrity Instagram Stories and public statements now hit the feeds with typos intact. Instead of embarrassment, the response is praise: evidence of someone speaking from the heart. It seems that even world leaders are not immune. Trump’s error-ridden posts on platforms like Truth Social are, according to his press office, evidence of his “excellence.” The Atlantic asked out loud: are typos the new status symbol?
The logical endpoint arrived in April of this year: Sinceerly, a Chrome extension built on Claude (yes, the AI) that deliberately introduces typos into your text. Presets run from “Subtle” to “CEO” (think the Anti-Grammarly). You use an AI to make your AI-assisted writing look human.
This is the new, utterly bizarre arms race we built ourselves into. Language used to be evidence of a human mind. Now, not so much. The question is how long typos remain useful as an indicator of authenticity before they get gamed out of meaning too.
The eyeball people
Sam Altman (Co-Founder and CEO of OpenAI) has a different proposal. Skip the typos (plus the CAPTCHAs, the proxies, and so on). Go straight to the source.
World, the project Altman co-founded (formerly Worldcoin, before the rebrand), operates a network of orb-shaped eye scanners. You look into the orb, and the thing scans and hashes your iris. You receive a World ID: a cryptographic credential proving you are a unique, living human being. There’s no name attached or entry that then into some government database. Just the proof of your biological distinctness, encoded.
So far, eighteen million people have done this across 160 countries. As of this spring, Tinder offers a “proof of humanity” badge for iris-verified users. Zoom and DocuSign have integrated it. OpenAI is reportedly building an entirely humans-only social network that would use either Orb verification or Apple’s Face ID to gate access.
I understand lots of the objections. The surveillance implications are very real. The early-adopter token distribution (users received WLD cryptocurrency for scanning) attracted regulators across Europe and Asia. And there is something viscerally uncomfortable about a company, chaired by the CEO of the world’s most influential AI lab, becoming the arbiter of who counts as a real human online.
Those objections are totally legitimate. And so is the problem they are trying to solve.
CAPTCHAs, for example, have largely become a joke. Bots now pass them more reliably than humans. AI-generated documents fool identity systems designed in a different/bygone era. The traditional gatekeeping infrastructure was built for a world where bots were primitive and rare. That world is completely gone.
Why the blockchain argument is different this time
For years, crypto advocates promised that decentralized trust would cut out the middleman across basically every industry. Mostly, what happened in the early days was: NFTs (remember those knuckleheads paying 6-figures for pictures of rocks and monkeys) and some fraud.
But the underlying proposition for blockchain identity makes a lot more sense than it ever did for currency speculation. And the problem it’s solving is more urgent, given the current proliferation of AI and bots.
The core challenge is what cryptographers call the Sybil attack: one entity creating thousands of fake identities. Dictators, bot farms, and AI agents all exploit the same vulnerability. Government IDs verify identity, not uniqueness. FaceID confirms who you are, not that there’s only one of you.
What you need is proof that this account maps to exactly one living human, without creating a central database of which human.
Zero-knowledge (ZK) proofs do just this. A ZK proof lets you demonstrate a property (I am a unique human) without revealing the underlying data (exactly which human). The verification is legit and importantly, the surveillance is optional. It’s the cryptographic equivalent of proving you’re old enough to drink without handing over your driver’s license (with all that other personal information attached to it).
Humanity Protocol uses palm biometrics plus ZK proofs. Academic researchers at Berkeley published a formal cryptographic framework for proofs of personhood in February. The Linux Foundation and Hedera ran a working demo in March. World launched AgentKit with Coinbase to let AI agents carry cryptographic proof of a human principal (i.e. every agentic transaction traced back to an accountable person).
The plumbing is being built and the whole crypto industry has matured significantly in recent years. This is less exciting than a new social feature, but more important than most things shipping in tech right now.
Side note: If you are looking for a no-BS, deep-dive into the world of Blockchain and the future of the internet, I suggest checking out the book “Read Write Own: Building the Next Era of the Internet” by Chris Dixon. It argues that the internet is entering a new “read-write-own” era, powered by blockchain technology, which can reverse the consolidation of power in the current era (dominated by large corporations) and restore the open, decentralized ideals of the early internet.
The internet that’s coming
A verified-human internet could look like this.
Platforms require proof of personhood for high-trust events (e.g. voting in polls, leaving reviews, asking for emotional support, going on dates, etc.). AI can still participate, but it must identify itself. The ambiguity that defines the current moment gets resolved, one credential at a time.
Reddit is moving in this direction. It now requires accounts flagged as suspicious to pass human verification before posting. Digg, which didn’t act fast enough, just shut down. The incentive structure is becoming clear: platforms that can credibly guarantee human users are worth more to people who want genuine human interaction. Which is most of the things worth doing online.
The dating app use case offers a strong perspective into this topic right now. Tinder’s proof-of-humanity badge is, at the very least, a promise: the person you’re swiping on exists. That is now a premium feature (which, alas, is it’s own kind of problem within the current approach to building tech). We built a global infrastructure for human connection and ended up at a place where actual human beings come with a verification badge.
The governance problem: whoever controls the proof of humanity controls the internet.
This doesn’t get resolved just by building better technology.
Whoever issues proof of personhood credentials controls, in some real sense, who gets to fully participate in the digital world. World is a private company (which Altman chairs). His other company is OpenAI, which is building the AI that most threatens to make human-ness irrelevant online. He is on both sides of the problem simultaneously.
Open-source alternatives do exist, however. The Linux Foundation work is a genuine effort at infrastructure without corporate ownership. Proof-of-Humanity, the earlier Ethereum-based project, attempted a fully decentralized registry. But, none of them have World’s scale.
Returning to the humble typo: it’s a workaround for a broken assumption. One that says text alone tells you who is on the other side of the screen. It really doesn’t anymore. We are rebuilding the trust layer of the internet in real time, under pressure, with inadequate governance frameworks and a handful of competing private interests racing to become the default.
The stakes (specifically, who controls the proof of humanity) are enormous, and being mostly discussed in academic cryptography papers and startup pitches rather than in any mainstream, public forum that resembles democratic discourse.
A typo bought us a little time. Not much more.
P.S. How many typos did you find in this article?
Malin King is the co-founder of Fynch: a fully white-labeled AI platform built by marketers, for agencies. AI + Human Experts. Learn More at getfynch.com.