
The internet is losing the ability to know who is human.
AI agents can now browse websites, fill out forms, complete tasks, and produce responses that are increasingly difficult to distinguish from those of real people. The systems the internet has historically relied on to separate humans from machines were not built for this world.
The Proof of Human API continuously distinguishes humans from bots and AI agents.
Modern AI can produce very human-like outputs. But producing an answer and producing it like a human are different things. Humans hesitate. We make mistakes. We correct ourselves. Our attention changes over time. The way we type, move, and interact leaves rich signatures that remain difficult for AI systems to reproduce. We call the broader research program the Invisible Turing Test for the Internet.
Proof of Human runs in the background while someone uses a website—no CAPTCHA, no puzzle, no interruption. We combine behavioral, device, and network signals and return a real-time assessment that companies can use to decide whether to trust an account, accept a response, or pay for work.
https://www.youtube.com/watch?v=swJabsQubf4
We tested scripted bots and AI agents against Proof of Human and other leading detection systems.
On a five-question survey, Proof of Human identified 99% of automated sessions, followed by reCAPTCHA v3 at 69.4%, hCaptcha at 44.3%, Cloudflare Turnstile at 42.1%, and Cloudflare Bot Management at 9.1%.
Traditional automation signals remain effective against scripted browsers, but they are less reliable when AI agents operate through genuine browser environments. Even interactive challenges may offer limited protection when vision-capable agents can interpret and complete them. As browser-level indicators become easier to avoid, the way a user behaves throughout a session becomes increasingly important for distinguishing human activity from automation.
Today, Proof of Human acts as the trust layer behind online human research. Stanford Psychology uses it to verify that participants in online studies are real. GrowProgress uses it as part of its quality-control stack for polling and public-opinion research. Market insights firms including Obsurvant use it to identify AI agents and other fraudulent respondents before they enter client datasets.
The common problem is the same: a response itself is no longer enough. An AI agent can give a coherent answer, pass an attention check, and look perfectly legitimate in the final dataset. Proof of Human looks at the process that produced it—how someone types, clicks, scrolls, moves through a page, and interacts with their device—continuously and without interrupting the user.
Production traffic also teaches us what to look for next. Customer deployments have uncovered new patterns of automation, from unnatural typing to partial AI copy-pasting, that are difficult or impossible to identify from the final answer alone. Those behaviors become new detectors, which are then tested across human, bot, and agent traffic before being deployed back into the network. The result is a system that improves as the adversary changes.
Proof of Human at Stanford Psychology. We first encountered this problem as researchers ourselves; today, research groups use Proof of Human to verify that the people behind online studies are actually people.
We started with online research because the problem was acute and measurable: organizations were paying for human judgment without reliably knowing whether a human had produced it.
But the same problem is rapidly appearing across applications, marketplaces, transactions, social platforms, and digital work. The internet needs a way to know when a human is actually on the other side—continuously, invisibly, and without another CAPTCHA.
Welcome to the Proof of Human era.
Matt & Mayank