CAPTCHAs can still detect AI agents

TEXT ANALYSIS: Roundtable Technologies / Proof of Human

The Dissection

This is a sophisticated sales deck with academic credentials. The authors present a technical framework—Process Turing Test, CogCAPTCHA30, cognitive process fingerprinting—as proof-of-concept for "Proof of Human," their commercial authentication product. The structure is textbook academic marketing: real findings embedded in a narrative that serves the startup pitch.

The Core Fallacy

The paper conflates verification with value preservation. Under DT logic, these are completely different problems. The authors assume that proving "this is a human, not an AI" will preserve something economically meaningful. But the DT thesis states that post-WWII capitalism dies because AI severs the mass employment → wage → consumption circuit—the circuit doesn't care whether you can prove you're human. If you're unemployed because AI replaced your cognitive labor, your "Proof of Human" certificate is worthless.

They're solving a gatekeeping problem while the participation problem renders gates irrelevant.

Hidden Assumptions

1. Humanness will remain economically scarce. The paper assumes that being verified as human provides access to something AI cannot replicate. But this assumes the economic value of human-only domains will persist at scale. DT says it won't.

2. Verification moats are durable. The paper's own findings demolish this: "when given full information—the observed features and the discriminator's objective function—the gap between humans and agents disappears." Any moat here is temporary, exploitable, and incentive-aligned for AI developers to close.

3. Cognitive process ≠ economic value. The paper treats humanlike process as the discriminating variable. But under DT, the relevant variable is productive participation, not cognitive signature. An AI that doesn't think like a human but produces all economic value is still catastrophic for human economic participation.

Social Function

Prestige signaling + transition management theater. The Princeton and Berkeley PhDs lend institutional credibility. The framing—"humans are still distinguishable from AI, here's how"—offers a comforting narrative that there's still a human-machine boundary worth defending. This soothes enterprise buyers and investors while the underlying displacement accelerates.

The Verdict

The most telling passage: "while frontier models are becoming more powerful over time, they are not necessarily becoming more human. Contemporary progress in artificial intelligence is independent of progress in human simulation."

This is the DT thesis stated plainly by people who don't understand what they've found. Capability and humanness are diverging. The market will reward capability. The verification layer will become a cat-and-mouse game that AI wins by default as stakes rise and incentives sharpen. The authors are selling locks to people who should be asking why the walls exist at all.

Structural Assessment: This paper describes a temporary moat at the authentication layer while the economic participation layer—the one that actually matters for human survival in post-WWII capitalism—is being dismantled by forces this analysis doesn't even gesture at.

Bottom line: They're building better padlocks for a sinking ship.