The Label Tax: An Autopsy of AI-Washing in Crypto Media

We didn't get a story. We got a press release wearing a story's clothes, and somebody stapled a "Web3" tag to its collar before shipping it downstream.

Here is the entire factual payload of the item I spent a full day dissecting: a bank and an unidentified partner are building an "AI deposit onboarding platform." That's it. One title fact. Four sentences of marketing residue. No token. No chain. No wallet. No TVL. No gas. No funding round. No valuation. No code repository. No audit. No timeline. No quantitative target of any kind. Five information points total, and one of them is the headline itself.

And yet it was filed under crypto. And yet it will be consumed as crypto. And yet, by the time you finish reading this, some fund somewhere will have added it to a "AI x Crypto momentum" watchlist without a single on-chain byte to justify it.

That is not a news problem. That is a taxonomy problem. And taxonomy problems are how bear markets quietly liquidate the people who trusted the labels instead of the ledgers.

Context: What We Are Actually Looking At

Let me be precise, because precision is the only thing that survives a bear market. The subject of this autopsy is a short item published on a crypto-native outlet describing a collaboration between Customers Bank — an American chartered depository institution, part of Customers Bancorp, which trades on Nasdaq — and an unnamed counterparty whose identity was lost somewhere in the pipeline and now exists only as a bracketed placeholder.

The product being announced: an AI-driven "deposit onboarding" system. In plain language, this is the process by which a customer opens an account, verifies identity, and passes source-of-funds review so that money can legally land in the vault. The core machinery of that process is KYC and AML — Know Your Customer and Anti-Money Laundering compliance. The AI, whatever it is, almost certainly sits on top of that flow, automating identity resolution, sanctions screening, risk scoring, and the endless paperwork that makes bankers weep.

This is not a blockchain. It is not an L1 or an L2. It is not a rollup, a bridge, a sequencer, or a validator set. It is a business-process automation layer bolted onto a traditional bank's customer intake funnel. The technical category is application-layer FinTech, and if you want to be generous, RegTech — regulatory technology.

The only thing connecting this story to crypto is that a crypto outlet published it, and that one of the named parties happens to have a history in the digital-asset banking space. The web3-ness of this article is nominal, not substantive. It is a costume. And I want to spend the rest of this piece explaining why costumes matter, how they get manufactured, and how you stop paying for them.

First, a definition that the industry has been slow to formalize. You know "greenwashing" — the corporate art of draping a coal plant in recycled paper and calling it sustainability. AI-washing is the same move with a different word on the banner. It is the practice of attaching an AI label to a product that either predates modern machine learning, or uses commodity models as a marketing garnish, or exists primarily so that a press release can contain the letters A and I in that order. In 2023 through 2025, as AI became the single dominant narrative in global capital markets, AI-washing went from a curiosity to an epidemic. And crypto media, which lives on narrative velocity, became its most efficient distribution channel.

The Forensic Teardown: Reading a Story That Isn't There

I have done this before. In 2017, I spent a full day with the Golem network's pre-sale contracts, reading distribution logic line by line with a pen and a legal pad, and found three flaws in the token emission algorithm that could have inflated supply past any defensible ceiling. I filed the issue, the protocol paused, and I learned a lesson that has never stopped paying: the story a project tells you and the story its code tells you are two different documents, and only one of them is sworn testimony.

So when I sat down with this item, I didn't read it for meaning. I read it for mass. I counted the verifiable facts. I subtracted the adjectives and weighed what was left. What was left was almost nothing, and the shape of that nothing is the actual news.

Let me show you the arithmetic. The item contained five discrete information points. One: the headline fact, that two parties are collaborating on an AI deposit platform. Two through four: three variations on the theme of "this will revolutionize banking efficiency," "this sets a new industry standard," and "this opens a channel to commercialize the product." Five: the outlet itself. That is the whole skeleton.

Now apply the first filter any serious reader should own — the falsifiability test. What claim in this article could be proven wrong? "Revolutionizes banking efficiency" cannot be falsified because no baseline, no metric, and no timeframe are supplied. "Sets a new industry standard" cannot be falsified because no standard is named. "Opens a commercialization channel" cannot be falsified because no contract value, no revenue split, and no customer count is disclosed. An article in which every substantive claim is unfalsifiable is not an article. It is an advertisement with a byline.

This is where the pseudocode earns its keep. I write classifiers for media pipelines the same way I write them for contracts, and the logic that should have caught this is embarrassingly simple:

def classify(article):
    web3_features = ["chain", "token", "wallet", "defi", "l1", "l2",
                     "rollup", "tvl", "gas", "validator", "bridge", "mempool"]
    score = 0
    for term in article.body:
        if term in web3_features:
            score += 1
    if score == 0:
        return "FinTech"          # default to the boring truth
    if article.source in crypto_outlets and score == 0:
        return "NON_NATIVE — flag for narrative pollution"
    return "Web3"

Run this item through that function and it returns a hard zero. No chain. No token. No wallet. No DeFi. No L1. No L2. No rollup. No TVL. No gas. The classifier says FinTech, and the classifier is right. The only reason this thing ever wore a Web3 tag is that a crypto outlet is a crypto outlet, and newsroom defaults are lazy. Source-based classification is a category error, and it is the single most common form of narrative pollution in this industry.

I want to name the mechanism precisely, because I have watched it destroy more capital than any exploit. I call it the Label Tax. It works like this. Every asset, story, and protocol in crypto carries a label. The label is not a description; it is a claim on your attention. When a story with zero on-chain substance is filed under Web3, it doesn't just occupy a slot — it borrows the credibility of everything genuinely native to that slot. The reader's brain, which is a lazy Bayesian, updates: "Web3 story, therefore Web3 relevance, therefore possibly Web3 upside." The label did the work. The content didn't. And you paid the tax in misallocated attention, which in a bear market is the only currency that compounds.

Behavioral Resonance Mapping: Why the AI Label Still Works

Here's the uncomfortable part. The AI label works not because the AI is real, but because the audience is primed. In 2021, I built something I called a Resonance Index to quantify exactly this phenomenon for the Bored Ape Yacht Club. I ignored floor prices entirely. I mapped social capital — who owned what, who was photographed with whom, which holders conferred status by association. The index told me the market would peak weeks before it did, and it told a handful of angels to exit while the exits were still wide. The lesson was not about apes. The lesson was that in narrative markets, the emotional undercurrent moves before the price, and the label is the surface where that undercurrent becomes visible.

So let me map the AI narrative's resonance honestly. Since early 2023, AI has been the global market's main character. It is the one story that unites retail, institutions, and sovereign funds in a single conviction trade. That dominance creates a gravitational field. Anything that touches AI gets pulled toward the light. A bank automating its KYC flow is, in substance, doing something banks have done since 2016 — progressive digitization of compliance. But dress it in the word "AI" in 2024, and it inherits a sliver of the same narrative mass that moved Nvidia.

This is where the math gets ugly, and where I start to enjoy myself. The social-to-fundamental heat ratio on this item is north of five to one. Five units of narrative heat for every one unit of verifiable substance. I have measured this ratio across four cycles now, and it has never once been a buy signal. When the ratio crosses three, you are in late-stage narrative inflation. When it crosses five, you are in the zone where the label is doing all the work and the underlying is doing none.

Compare it, for a moment, to a genuinely native narrative. Take Bitcoin's Ordinals wave. Whatever you think of monkey JPEGs on the oldest chain, the inscription phenomenon did something measurable: it pushed real fee revenue into the block space and gave miners a second income stream at precisely the moment the emission schedule made the security budget a live question. That is a narrative attached to a mechanical consequence. Fees were paid. Hashrate was defended. The story and the ledger agreed with each other.

The deposit-onboarding story has no mechanical consequence anywhere. There is no fee, no block, no emission, no ledger. There is a slogan and a bracketed placeholder where a company name should be. One of these is a narrative with a body. The other is a narrative with a voice and nothing else.

And here is the resonance insight that should make you cold: the AI-washing item works on the same psychological lever that every successful crypto narrative works on — the fear of missing the next paradigm. In 2020 I spent two weeks modeling Uniswap V2's geometric mean pricing and concluded that the real revolution wasn't yield, it was permissionless liquidity. That conclusion was correct and it made me money. But notice what made it correct: there was a mechanism, a constant-product formula, a mathematically provable property. The AI story here has no formula. It has a feeling. And feelings, unlike formulas, cannot be verified — which is precisely why they are so easy to sell.

Narrative Decay Audit: Terra as the Control Group

I keep a graveyard. It's not morbid; it's methodological. Every cycle produces a set of stories that were believed, priced, and then dismembered. Reading the bones tells you more about the living than any bull-case deck ever will. The reference corpse for label fraud is Terra/Luna, and I spent three months of 2022 dissecting its mechanism for a ten-thousand-word piece I called "The Mathematics of Delusion."

The Terra case is instructive here not because it was a fraud — that debate is settled — but because of how long the label outlived the substance. For eighteen months, UST carried the label "algorithmic stablecoin." The label was technically true and economically suicidal. The mechanism required perpetual growth to hold the peg, which meant the narrative required perpetual believers, which meant the label required perpetual reinforcement. When the reinforcement stopped, the mechanism didn't just weaken. It inverted. The bug wasn't in the oracle. The bug wasn't in the code at all. The bug was in the story, which had quietly become load-bearing.

The lesson I extracted and now apply to every item I read is this: a label is dangerous in exact proportion to how much weight it carries and how little mechanism it describes. Terra's label carried enormous weight and described a mechanism that could not survive a bad week. The AI-washing item's label carries modest weight and describes no mechanism at all. Different pathologies, same family. Both are cases where the story is doing labor that only a ledger should be allowed to do.

Now run the same audit on the customers-bank item and watch the decay curve flatten into nothing. What happens to this narrative in six months? Two paths. Path one: quantitative results emerge — "deposit processing time cut by X percent," "false-positive AML flags reduced by Y percent." If that happens, the story earns a body and graduates from narrative to fact. Path two, which I assign the higher probability: silence. No follow-up, no metrics, no disclosure. The item joins the great sediment layer of AI-collaboration announcements that dissolved into the archive without ever touching a balance sheet. And the label tax was paid by everyone who read it as signal.

I have watched this sediment accumulate for years. In 2025, consulting for three Swiss banks entering the space, I did the opposite of what this article does. I synthesized fragmented crypto narratives into a coherent institutional adoption story that emphasized stability and compliance over decentralization purity. I argued to those banks — against the instincts of every crypto native in the room — that mass adoption requires narrative dilution. You cannot sell decentralization to a risk committee. You sell them predictability, and you let the label soften. That was a deliberate, honest dilution, built on real regulatory frameworks and real balance sheets. What we have in the customers-bank item is dilution without the substance — the costume without the body. Dilution is a strategy when it's backed by mechanism. It's a con when it isn't. The difference is whether the label points at a ledger or at a feeling.

The Economics of the Laundromat: Why This Keeps Happening

You cannot fix a pathology you refuse to model. So let me model the incentive structure, because the AI-washing item is not an accident. It is an equilibrium.

Crypto media runs on a simple loop: attention in, advertising and subscriptions out. In a bull market, attention is abundant and cheap — every price tick is a story. In a bear market, attention is scarce and expensive, and the outlets that survived the purge need volume. Where does volume come from when the market is quiet? It comes from adjacent narratives. AI is adjacent. FinTech is adjacent. Anything with a corporate logo and a forward-looking verb is adjacent. The outlet doesn't classify this item as Web3 because it believes it is Web3. It classifies it as Web3 because the pipeline runs on source, not on substance, and because a story is a story.

The company, meanwhile, has its own incentives. A chartered bank wants to be seen as innovative without taking innovation risk. Announcing an AI partnership is cheap, low-commitment, and optically modern. If the collaboration is a framework or a memorandum rather than a signed revenue contract — and the total absence of financial terms strongly suggests exactly that — then the announcement is the product. The press release is not a report on the business. The press release is the business, or at least a meaningful fraction of it.

This is the AI-washing laundromat: a company with a modest internal project, a media outlet hungry for adjacent volume, and a reader primed by a dominant narrative. Run them through the loop and out comes a "Web3 story" with the on-chain content of a grocery receipt. Nobody in the chain is lying, exactly. That's what makes it durable. Everyone is just doing their local job under a label that nobody audited.

I've been on the other side of this loop, and I know how seductive it is. When my 2020 Uniswap piece went viral, I got invitations to closed hackathons and private rooms where the incentives to keep the narrative hot were enormous. The discipline that kept me honest was mechanical, not moral: I refused to publish a claim I couldn't attach to a number. It cost me reach. It saved me credibility. In a market that remembers everything and forgives nothing, that trade is always correct.

And note where the loop breaks. It breaks at the reader who insists on mechanism. If you had demanded, of this article, a single number — deposits processed, false positives reduced, contracts signed, revenue booked — the entire narrative would have collapsed into the empty set. The laundromat only works on readers who don't ask for the invoice.

The Real Signal: Customers Bank and the Compliance Ledger

Beneath the marketing fog, there is exactly one thread in this item with genuine analytical weight, and it has nothing to do with AI and everything to do with regulation. The named party, Customers Bank, is not a neutral bystander in the crypto banking wars. It is a chartered American institution with a public history in the digital-asset banking space, and that history is the only place where this story touches something that can actually move.

Publicly, Customers Bancorp and its bank subsidiary drew regulatory attention over risk-management and anti-money-laundering deficiencies, with an enforcement posture that put the institution under formal scrutiny. I flag the confidence level honestly: this is reconstructed from public memory and should be independently verified by the reader. But the analytical point survives regardless of the exact paperwork. When a bank under compliance scrutiny announces an AI-driven onboarding and screening platform, the honest reading is not "innovation." The honest reading is "remediation wearing innovation's jacket."

This reframes the entire item. The AI isn't there to grow revenue. The AI is there to answer a regulator. Automated KYC and AML screening directly addresses the deficiency that enforcement actions target, and demonstrating compliance-technology investment is exactly what an institution does when it needs to rebuild supervisory trust. That is a real, defensible corporate motive. It is also completely disconnected from the crypto market's price action, which is why the item's true market impact is close to zero.

And it introduces a second-order risk that the marketing never mentions. Using AI for anti-money-laundering screening is itself a regulatory minefield — model explainability, false-positive rates, potential bias in risk scoring, the question of who is liable when the model makes the wrong call on a real customer. An AI that automates compliance is an AI that automates compliance risk. You cannot outsource judgment to a model and then disclaim the judgment. Code is law, but liquidity is truth — and here the liquidity is regulatory capital, which means the model's errors get priced in dollars, not in gas.

Here's the part that should sharpen your attention. If any slice of that deposit onboarding touches crypto-native customers — and a bank with this particular history might well have them — then the item acquires a faint, indirect Web3 relevance after all. But notice the chain of inference required to get there: unnamed partner, unstated customers, unverified business lines, inferred motive. That is not analysis. That is speculation in a lab coat. And speculation in a lab coat is the most expensive costume in the industry.

The Detection Framework: How to Stop Paying the Label Tax

I don't write autopsy reports for the pleasure of it. I write them so the next corpse is cheaper. So here is the framework I now run on every crypto-labeled item, and I offer it in the same spirit I offered that Golem issue in 2017: not to embarrass the target, but to install a check that stops the error from compounding.

Filter one: the feature-word census. Count the on-chain vocabulary. Chain, token, wallet, contract, TVL, gas, validator, bridge, mempool, rollup. If the count is zero, the item is not crypto. Full stop. The source doesn't matter. The byline doesn't matter. The outlet's brand doesn't matter. A crypto outlet publishing a non-crypto story is a non-crypto story.

Filter two: the falsifiability test. For every substantive claim, ask what observation would prove it false. "Revolutionizes efficiency" fails. "Sets a new standard" fails. "Opens a commercialization channel" fails. A story composed entirely of unfalsifiable claims is a story composed entirely of marketing.

Filter three: the mechanism audit. Every legitimate crypto narrative has a mechanical consequence. Ordinals pushed real fees to miners. Uniswap V2's constant-product formula provably made liquidity permissionless. Post-Dencun blob space is a finite resource that will saturate — and when it does, every rollup that has been subsidizing cheap transactions on cheap data will watch its gas costs climb, because the discount was never structural, it was a temporary surplus. If you can't name the mechanism, you're not holding a thesis. You're holding a vibe.

Filter four: the incentive trace. Who benefits from this item existing in this form? Follow the press release back to the balance sheet. If the announcement is the product, treat it as advertising, not disclosure.

Run all four filters and the customers-bank item scores zero on every one. That is not a judgment about the bank. It is a judgment about the label. And labels are the only thing I'm allowed to audit from the outside, because the inside is where the unpaid invoices live.

One more filter, and it's the one that costs the most to apply. Filter five: the discomfort test. If a story makes you feel smart for understanding it and excited for holding it, and you cannot point to a single number behind it, the feeling is the tell. Narrative markets weaponize exactly this feeling. The staccato thrill of being early, the asymmetric rush of a label you recognize — these are not signals. They are the tax collector's hand on your shoulder.

The Contrarian Angle: The Label Isn't Wrong — It's Dying

Everyone in this industry wants to blame the mislabeling. I'm going to argue the opposite, and I expect it to be unpopular.

The real story here is not that a FinTech PR piece got filed under Web3. The real story is that "Web3" as a category is being absorbed, in real time, by "AI" — and the mislabeling is the symptom, not the disease. Watch the vocabulary migrate. In 2021, everything was "decentralized." In 2023, everything was "modular." In 2024 and 2025, everything is "AI." The industry's native narratives — ZK, L2, RWA, DePIN — are losing the battle for mindshare to a general-purpose story that doesn't even need crypto to exist. The AI-washing item is not an intruder in crypto media. It is the future arriving early. It is a preview of a world where the crypto label stops commanding a premium and starts merely borrowing one.

This should terrify the pure-play maximalists and quietly delight anyone who thinks in mechanisms. Because if the label stops paying, the only thing left is substance. The protocols with real fee revenue survive. The ones whose "utility" was a slide deck don't. I've said for years that liquidity mining APYs are just a project subsidizing its own TVL number — pull the incentives and the "users" evaporate, because they were never users, they were yield tourists following the highest number. The AI-washing phenomenon is the same disease in the narrative layer: a story subsidizing a label it didn't earn, and the subsidy is about to run out.

So here's the contrarian read. The label tax is not a bug in crypto media. It is the mechanism by which the industry discovers which labels were never worth the tax to begin with. Every mislabeled item is a stress test. The labels that survive a bear market of mislabeling are the ones attached to real ledgers, real fees, real mechanisms. Everything else was a costume, and bear markets strip costumes for a living.

Takeaway

The next time a story crosses your desk wearing a Web3 tag, don't read the tag. Read the ledger. Count the feature words. Demand the invoice. And when a label promises you a paradigm while its content offers you a feeling, remember what the graveyard already knows: the labels decay first, the mechanisms decay last, and the difference between the two is the only edge that survives a bear market.

We didn't get a story this time. But we got the receipt for how the next one will be manufactured — and that receipt, unlike the press release, is worth keeping.

Code is law. Liquidity is truth. Labels are neither — until you make them prove it.