Could a One-Won Stamp on Every Email Kill Spam? · Issue #41

Opening

Hello, dear reader. Let me start with a question. Have you ever made an international call back in the 1990s? I remember the way people would jot down exactly what they wanted to say before picking up the phone, because the meter was ticking at hundreds of won per minute. Back then, we knew in our bones that communication had a cost.

What about now? Sending a single email costs, in practice, nothing. Same for texts, same for messaging apps. But the landscape this “free-ness” has produced is strange. As of 2025, roughly 376 billion emails circulate the globe every day, and nearly half of them are spam. Even more striking: a majority of that spam is now written by AI.

Here’s the core of today’s story. When sending a message costs zero, the recipient’s attention becomes a zero-cost resource too. And what it might take to reverse this could be surprisingly simple.

Free Messages Have Created a “Tragedy of the Digital Commons”

Economics has a concept called the “tragedy of the commons.”¹ On an open pasture shared by everyone, if each herder grazes as many cattle as possible, the grass eventually disappears and no one can use it anymore.

The email inbox follows exactly this structure. Since sending costs the sender nothing, it’s rational for marketers and spammers alike to send as much as possible. As a result, the recipient’s inbox has become a pasture stripped bare.

Korea’s situation is no different. According to a joint survey by the Korea Communications Commission and KISA (Korea Internet & Security Agency), the average monthly volume of illegal spam received per person hit a record high of 16.34 messages in the first half of 2024. After the government rolled out a strong comprehensive countermeasure, that number dropped to 11.60 in the second half — but the data for the first half of 2025 reveals an even more interesting pattern. Text-message spam plunged 58.5% to 3.04 messages per month, while voice spam actually rose about 40% to 2.13 messages per month.

Lock the door, and they come through the window. Economists call this the “balloon effect” — as long as the zero-cost structure of sending remains unchanged, spam simply switches channels and keeps going.

AI Made the Pasture “Infinite”

Then AI entered the picture, and things got a notch worse.

According to a joint study by Columbia University, the University of Chicago, and Barracuda, 51% of spam emails as of April 2025 were AI-generated. The research team set emails sent before the public release of ChatGPT (November 2022) as the “human-written” baseline, then estimated the share of AI usage by comparing subsequent data against it. What’s interesting is the character of AI-written spam. It has fewer grammatical errors, reads more formally, and is more linguistically polished. In other words, it’s gotten easier to slip past spam filters.

Phishing² is even more dramatic. In experiments run by the cybersecurity firm Hoxhunt, AI spear-phishing agents³ performed 31% worse than human experts in 2023 — but by March 2025, they were performing 24% better. The tables turned in just two years. What’s playing out right now is essentially an arms race between AI filters and AI spam. Both defense and offense are running on AI. But there’s a structural asymmetry baked into this arms race: the attacker only needs one message to succeed, while the defender has to stop every single one.

The “Digital Stamp” — Why a 20-Year-Old Idea Failed

Actually, a solution to this problem was proposed 20 years ago: “impose a tiny cost on every email.”

In 2003, Microsoft Research announced the Penny Black Project. The name is telling — it’s a nod to the Penny Black, the world’s first postage stamp, issued in Britain in 1840, which shifted the cost of mail from the recipient to the sender. The idea behind the Penny Black Project was simple: make the sender’s computer burn a certain amount of computational resources every time it sends an email. This has zero impact on an individual sending one or two messages, but becomes a crippling cost for a spammer who needs to send millions.

Similar attempts followed — Yahoo’s CentMail, and several Hashcash-based projects. All of them failed. MIT Technology Review had already flagged the core problem back in 2004: ultimately, this was never a technology problem. Who operates the system? Who bears legal liability? How do you build a global standard? These are questions of institutions and incentives, not protocols — and that’s why the project ultimately failed.

And above all, users already accustomed to “free” simply wouldn’t accept any form of cost. For what it’s worth, my own newsletter uses Maily to give you quality content and a good experience, and it costs roughly ₩1,000,000 (~$720) a year to run. Please, more paid memberships… (I’m running a serious deficit).

What the Peacock’s Tail Tells Us: Costly Signaling Theory

Let’s step back for a moment and look more fundamentally at why “free signals” are a problem.

Biology has a theory called Zahavi’s Handicap Principle.⁴ A male peacock’s flamboyant tail is a survival liability. It’s heavy, it draws attention, and it makes escape harder. But that very “liability” is the point of the signal. It proves “I am healthy enough to bear this handicap.” Being able to afford the cost is itself evidence of honesty. In economics, Michael Spence’s signaling theory⁵ follows the same logic. A college diploma doesn’t necessarily make someone more productive, but the signal that “I bore four years of time and expense” has real value to an employer.

There’s even experimental proof of this. In a 2019 study published in PNAS by Tchernichovski and colleagues, researchers imposed a tiny time cost on an online rating system — a slider that took longer to drag all the way — and rating accuracy improved significantly. People who had rated carelessly when it was free became noticeably more deliberate once even a small amount of effort was required.

The same logic applies to a one-won email stamp. That one won isn’t really about money — it’s about courtesy. It signals, “I cared enough about your time to bear at least a minimal cost in sending this.” One won per email means nothing to an individual, but for a spammer sending 10 million messages a day, that’s ₩10,000,000 (~$7,200).

The Lesson from X (Twitter): Design Is Everything

So how have real-world attempts to price communication actually fared?

Elon Musk’s X is a fascinating test case. Paid verification (US$8/month) was meant to curb bots and spam, but the results were mixed. Some bots did disappear — but a paradox emerged where impersonation accounts that had purchased the paid checkmark actually gained more trust.

The lesson here is clear. The principle of “impose a cost” is sound, but “on whom, and how” is the entirety of the design problem. X didn’t impose a cost on the sender per message — it imposed a cost on maintaining the account. For a spammer, US$8 a month is just a business expense, not a “signal of courtesy” attached to each individual message.

An effective “digital stamp” system would need to whitelist messages from people you know or services you’ve subscribed to — free of charge — and impose a small fee only on unknown senders. It’s essentially the digital version of the cultural practice of buying a stranger a drink the first time you meet them at a bar.

X has also just launched a new experiment: a weekly payout system. It pays out ad revenue based on the views and reach of a post — and while existing platforms like YouTube and TikTok settle accounts monthly, X now settles weekly, letting everyone run all sorts of experiments. Of course… thanks to this, X is now awash in an endless carnival of ragebait and gimmicks, all chasing that weekly paycheck.

Oz’s Lens

Honestly, I don’t think the “digital stamp” is likely to become reality. There’s a reason it’s failed for 20 years running.

But it’s worth paying attention to why this idea keeps coming back. Something I’ve noticed constantly while building go-to-market strategies: the more fundamental the problem, the simpler the proposed solution tends to be, over and over. And the reason that solution keeps failing is, more often than not, incentive structures — not technology.

Here’s the situation we face today, laid out in data: a majority of spam is now AI-made, AI phishing has surpassed human experts, and in Korea, blocking text spam just pushes it to voice, and blocking voice just pushes it somewhere else. This is whack-a-mole. Plug one hole, and it pops up somewhere else.

To me, the real question isn’t technical blocking. It’s a question of institutional design: “In a world where communication costs nothing, how do we protect the resource called attention?” And this question extends well beyond spam — to social media algorithms, the digital advertising industry, and even the information environment of democracy itself.

Herbert Simon said it back in 1971: a wealth of information creates a poverty of something else. That something else is attention. Half a century later, we still haven’t built the institutions to protect it. A one-won stamp may not be the answer. But I believe the principle that “a tiny bit of friction guarantees the honesty of a signal” is an axis we’ll have to reckon with whenever we design digital communication going forward.

Closing

Here’s today’s story in three lines. As the cost of communication dropped to zero, the recipient’s attention became a zero-cost resource too. AI is accelerating this dynamic, and filtering technology alone can’t win this arms race. The direction of the solution isn’t technical blocking — it’s institutional design that restores a small amount of “friction” back onto the sender.

Next time you open your inbox, try counting. How many of those messages actually respect your time? That ratio is a pretty good indicator of the health of our communication system right now.

References & Further Reading

Wei Hao et al., joint research by Barracuda/Columbia University/University of Chicago, “AI-generated spam” analysis, published 2025.06.18. : The core data source behind today’s claim that 51% of spam is AI-generated.
Hoxhunt, “AI-Powered Phishing Outperforms Elite Cybercriminals in 2025”, 2025.03. : Contains the experimental results showing AI phishing began outperforming human experts by 24%.
Ofer Tchernichovski, Lucas C. Parra, Daniel Fimiarz, Arnon Lotem, Dalton Conley, “Crowd wisdom enhanced by costly signaling in a virtual rating system”, PNAS, 2019, 116(15), 7256-7265. : An experiment showing that imposing a time cost on online ratings improved accuracy — the key paper empirically demonstrating a digital application of costly signaling theory.
Korea Communications Commission & KISA, “H2 2024 Spam Distribution Status,” published 2025.03.28. : Confirms trends in Korean spam volume and the effectiveness of government countermeasures.
Korea Broadcasting and Communications Commission & KISA, “H1 2025 Spam Distribution Status.” : The official report revealing the balloon effect — a 58.5% drop in text spam versus a 39.2% rise in voice spam.
Tim Wu, The Attention Merchants, Knopf, 2016. : Traces the history of the “attention economy” from 19th-century newspapers to modern digital advertising. Helpful for understanding the historical context of today’s topic. Published in Korea under the title “The Right Not to Pay Attention.”
Microsoft Research, Penny Black Project, 2003. : The original “digital stamp” project, and a case study in how a technical approach ran into institutional limits.

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis — business data management and business analytics — while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master’s from Korea University’s Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

Tragedy of the Commons: The phenomenon where an open resource shared by everyone is eventually depleted by individuals pursuing their own self-interest. Named by economist Garrett Hardin in 1968. ↩
Phishing: A cyber-fraud technique that impersonates a trusted institution to steal personal or financial information. ↩
Spear Phishing Agent: An AI system that automatically generates customized phishing attacks targeting a specific individual or organization. Far more sophisticated than generic phishing. ↩
Zahavi’s Handicap Principle: A theory proposed by biologist Amotz Zahavi holding that the costlier a signal is, the more honest the information it conveys. The peacock’s oversized tail is the classic example. ↩
Spence Signaling Theory: A theory by economist Michael Spence describing a mechanism whereby, under conditions of information asymmetry, one party incurs a cost to prove a trait about itself. Won the 2001 Nobel Prize in Economics for the analysis of educational credentials as a “signal” in the labor market. ↩