Mato just raised pre-seed Read more

Nate App: The AI That Was Just People

  • Jun 3, 2026
  • 20 min

Show notes

What the episode covers

Before there was AI, there was a man in a box — and the con has barely changed since 1770. This episode traces the fraud behind Nate, the AI checkout app that raised over $50 million from top venture firms while running on hundreds of human contractors in the Philippines completing every transaction by hand. Derek and Grant walk through how founder Albert Saniger built the pitch, staged the investor demos, and suppressed an internal dashboard that proved the automation rate was effectively zero from day one. Along the way, they connect the case to Wolfgang von Kempelen's 18th-century chess automaton, Amazon's Mechanical Turk, and a parallel modern case involving a literal human hidden inside a fast food robot. The DOJ and SEC filed charges against Saniger in April 2025; as of mid-2025, he was living in Barcelona and had still not been served. The entire scheme unraveled not because of investor oversight or regulatory monitoring, but because one journalist asked a hard question — which may say more about the current state of AI enforcement than any of the individual facts of the case.

Timeline

In this episode

8 moments worth skipping to. The timecodes match the player above.

  1. 0:15Introduction
  2. 2:16The Cabinet Has a Secret
  3. 5:05The Pattern Has a Name
  4. 7:47Buy in One Tap, Powered by Magic
  5. 10:19Standby Engineers and Staged Demos
  6. 12:45The Information Story That Ended It
  7. 15:30The Oldest Trick in the Box
  8. 18:34Outro

Quick answers

Straight from the episode

The questions this one settles, without the listen.

What was the Nate app and why did it turn out to be fraud?
Nate was a shopping app that raised over $50 million by promising AI-powered one-tap purchasing automation. According to DOJ charges, the automation rate was effectively zero from day one — the app relied on human workers completing transactions while the company presented it as autonomous AI technology.
How did the original Mechanical Turk con work and what does it have to do with AI startups?
Wolfgang von Kempelen's 1770 Mechanical Turk was a chess-playing machine that concealed a human master inside a cabinet, fooling audiences including Napoleon and Benjamin Franklin for 84 years. The episode uses it as the structural template for modern AI washing: show an impressive demo, hide the human doing the actual work, and control what investors are allowed to see.
What happened to Nate's CEO Albert Saniger after the fraud was exposed?
The DOJ and SEC filed parallel charges against Saniger in April 2025. As of mid-2025, Saniger is living in Barcelona and has not been served with the indictment. He has continued operating a venture capital firm while under indictment.
What role did staged demos and a suppressed dashboard play in Nate's fraud?
Nate conducted staged investor demos with engineers on standby to ensure they went smoothly. Saniger also built an internal dashboard that tracked the true automation rate — which showed the AI was not working — and then classified it as a trade secret to restrict access, effectively hiding proof of the fraud from investors.
What other AI companies have faced similar enforcement actions for exaggerating automation claims?
The episode traces an enforcement arc that includes Delphia, Global Predictions, and Presto Automation. Presto is described as a near-literal repeat of the Mechanical Turk structure: a fast food drive-through AI product that had humans operating it from behind the scenes while marketing it as autonomous.
What ultimately exposed Nate's fraud, and what does that reveal about investor due diligence?
A 2022 investigation by The Information exposed Nate's false automation claims, immediately killing a pending funding round. The episode frames this as deeply uncomfortable: the only failsafe against the fraud was a single journalist asking hard questions, while investors who had given over $50 million never looked closely enough to find the suppressed dashboard proving the AI did not work.

Transcript

The full conversation

Every word of the episode, 2,884 of them, in the order they were said.

Read the transcriptHide the transcript

DerekWelcome to Vaporware. I'm Derek, he's Grant, and today we are going all the way back to 1770.

GrantWhich, yes, is a weird place to start a tech podcast.

DerekSo get this, a Hungarian inventor named Wolfgang von Kempelen builds a chess-playing automaton, robes, turban, cabinet full of gears.

GrantWow.

DerekIt beats Napoleon, it beats Benjamin Franklin. For 84 years, audiences across Europe and America just believe. And there's a chess master hidden inside the cabinet the whole time.

Grant84 years?

Derek84 years, a guy in a box. That is the original AI washing.

GrantThe first demo was just a man pretending not to exist.

DerekBasically, and here's the thing, that exact structure, it shows up again in 2005 when Amazon names its crowdsourcing platform after it. And then it stops being a wink. It starts being a con.

GrantOkay, I'm with you.

DerekAlbert Saniger, Nate App, 2018. He pitches an AI checkout tool that can buy anything on the internet with a single tap.

GrantHow much did he raise?

DerekAccording to TechCrunch, over $50 million. Investors like Coatue, Forerunner, Renegade Partners.

Grant$50 million.

DerekAnd the DOJ says the automation rate was effectively... 0% from day one.

GrantWait, Wait, wait. Zero?

DerekZero. Hundreds of contractors in the Philippines manually completing every purchase. Just a guy in a cabinet updated for 2019.

GrantBold strategy.

DerekWe're going to walk through all of it, the staged demos, the suppressed dashboard, the personal stock sale, the whole architecture of the thing.

GrantAnd where is Nate now?

DerekBarcelona, as of mid-2025. Still not served. The case is just sitting there.

GrantOf course it is.

DerekOkay, let's start in 1770, SEGMENT one. In Vienna seventeen seventy Wolfgang von Kempelen wheels out a life sized wooden figure, a man in Ottoman robes seated behind a cabinet with a chessboard on top, and before a single piece moves he does something theatrical: he opens every door of the cabinet, holds a candle inside so the whole audience can see the gears and clockwork

Speaker 3Wow!

Derek; there is nowhere, he tells them, for a person to hide.

Speaker 4And the audience?

GrantBought that?

DerekCompletely. This is seventeen seventy Vienna, Grant-clockwork automatons were the cutting edge-inventors were building mechanical ducks that could literally digest grain-so when Kempelen enclosed those doors and the thing started moving chess pieces on its own people thought they were watching the future.

GrantOkay, so who did this machine actually beat? Like, how famous are we talking?

DerekOh, you know, Napoleon Bonaparte, Benjamin Franklin. Chess grandmaster Francois-Andre Philidor-Philidor actually won his match, but his son said afterward it was the most fatiguing game of chess his father ever played.

GrantNapoleon lost to a box.

DerekNapoleon tried to cheat first; made a deliberately illegal move; the machine moved the piece back; he tried again; machine removed the piece from the board entirely; third time, it swept every piece onto the floor.

GrantThe chess robot had more dignity than Napoleon!

DerekAccording to Chess.com, the machine toured Europe and America for eighty-four years—eighty-four years, Grant; then in eighteen fifty-four it burned in a fire. In three years later the owner's son finally published a full explanation of the secret.

GrantWait, the secret came out after it burned?

DerekYeah, inside the cabinet was a chess master on a sliding chair, following the moves by candlelight through a series of magnets under the board; levers controlled the arm; the gears were basically just decoration to fill up space.

GrantSo the transparency was the trick. He showed people the empty cabinet to

Speaker 5prove there was nothing behind it.

GrantAnd to prove there was nothing inside, but that inspection was choreographed, the person slid out of view each time a door opened.

DerekExactly; the proof of innocence was built into the performance; and here is the part that got me: in two thousand five Amazon named its crowd sourced human labor platform after this exact machine. Jeff Bezos literally called the workers "artificial artificial intelligence.

GrantSo the khan became an industry.

DerekAn industry! Which makes you wonder, if the same structure has been running for two hundred and fifty years, what does it actually look like when a modern company tries to pull it off? So here's the thing about that pattern: Amazon didn't just borrow the name as a cute reference. Bezos looked at the original Turk and basically said, yeah, that's a business model.

GrantWhich is a sentence I did not expect to say today.

DerekRight? In 2005, they launched a crowdsourced labor platform, named it Mechanical Turk, and Bezos himself called the workers artificial artificial intelligence. That's a direct quote.

GrantHold on, he said that out loud, like proudly?

DerekProudly. The whole joke was, hey, the machine isn't real, it's humans wink, except the workers were tagged by ID numbers, not names, and according to IEEE Spectrum, a twenty seventeen paper found they earned median wage of around two dollars an hour.

GrantTwo dollars an hour to be someone's invisible robot?

DerekTwo dollars an hour.

GrantWow.

DerekOnly four percent cleared the U.S. minimum wage. The platform even called the tasks human intelligence tasks, HITs.

GrantSo the facade of automation was the product. Amazon just sewed the cabinet.

DerekExactly it. And nobody called it fraud. That's what makes Amazon MTurk the hinge point here. The same structure, human hidden inside a machine. goes from parlor trick to investor pitch without ever crossing a legal line because Amazon was honest with the wink.

GrantOkay, so that's why I want to push on this. Transparent parlor trick is one thing, but when does the wink disappear and it just becomes lying?

DerekAnd that is the question that gets us to 2018.

GrantDo I want to know what happened in 2018?

DerekSo a guy named Albert Saniger founded a company called Nate Shopping App. The pitch was, you could buy anything on the internet with a single tap, no human

Speaker 5intervention.

DerekHuman involvement, pure AI.

GrantNo human involvement. He said those words?

DerekAccording to the DOJ indictment, yes. Neural networks, machine learning, the whole vocabulary. Fortune covered this earlier this year.

GrantHe raised over $50 million from serious investors.

Derek$50 million. Coatue, Forerunner Ventures, a $38 million Series A in 2021 alone. The pitch was airtight.

GrantExcept-

DerekExcept the AI couldn't actually complete a purchase. The DOJ says it never could, from day one.

GrantSo what was completing the purchases?

DerekThat's where it gets good. So picture the pitch room, 2019, peak AI mania, every deck has a machine learning slide, and Saniger walks in with a phrase that sounds almost too clean, Buy what you wish in a tap.

GrantI mean, that's a good line. I give him that.

DerekRight? And according to TechCrunch, he backed it with the full buzzword arsenal: AI, machine learning, neural networks, transactions completed (and I'm quoting the indictment here) without human involvement.

GrantOkay, and nobody asked for a live demo that they controlled?

DerekOh, there were demos. We'll get there in a second. But first, the money. TechCrunch reported Nate raised over $50 million. Coatue (Forerunner Ventures) a thirty eight million dollar Series A in two thousand twenty one led by Renegade Partners.

GrantThirty-eight million for an app whose AI, according to the DOJ, had an automation rate of effectively zero percent from day one?

DerekEffectively zero.

GrantWait, wait, wait—he bought the AI from a third party, right?

DerekMm

GrantDidn't

Derekhmm.

Granteven build it in house?

DerekThat's the part that gets me. Fortune reported the tech he purchased from a third party and I quote: Never achieved the ability to consistently complete e commerce purchases He had data scientists on staff He had a real team The technology just didn't work

GrantSo from day one he knew.

DerekAccording to the DOJ, yes, and here's where the Kempelen parallel snaps into focus. Saniger had

GrantWow.

Derekan actual internal dashboard showing the automation rate. He could see the number, zero, and, he told employees, the data was a trade secret. Nobody could look at it.

GrantA trade secret, the secret being that there was nothing there.

DerekExactly. And SC Media reported he told investors the rare campaign. Rare cases where humans stepped in were just edge cases, exceptions. The AI handles everything else.

GrantThe math doesn't work and he knew it. The question is how he kept thirty-eight million dollars worth of sophisticated investors from figuring that out.

DerekWell, turns out Saniger had engineers on stand by during every investor demo, manually completing test purchases in real time, making the app look fully automated.

GrantSo the demo was the cabinet.

DerekThe demo was the cabinet. Same trick, different century.

GrantOkay, I need to hear everything about those demos.

DerekSo the demo itself was the product.

GrantThe "demo" was the Turk. Open the cabinet, show everyone there's nothing inside, then close it and let the hidden guy do the work.

DerekExactly; and according to the SEC complaint, Saniger did this systematically. During investor product demos he directed Nate engineers to be on standby, manually completing test purchases in real time so

GrantWow.

Derekthe app looked fully automated.

GrantSo investors are watching this thing just fire off orders, seamlessly.

DerekRight—one tap, done. Feels like magic-and it was magic, just not the kind they meant.

GrantIn the least flattering sense of that word.

DerekNow here's the detail that I think is the most damning: Saniger didn't just hide the truth from investors-he built a dashboard, an actual internal tool that tracked the real automation rate.

GrantWait, back up. He built a tool that showed exactly how fake the product was?

DerekYes; and then the indictment alleges he restricted access to it, told employees the data was a trade secret, and directed staff not to share automation metrics with the rest of the company.

GrantSo the evidence that proved the fraud was classified.

DerekAs a trade secret, yes; the transparency was the con-same move Kempelen made with the candle tour, just with better documentation. you

Grantcouldn't be more careful about hiding something if you tried.

DerekIn the Philippines and Romania, teams were told to prioritize transactions initiated by investors. So during a fundraising demo, real humans were jumping the queue to make sure your test purchase went through instantly. He triaged the fraud. That's the word for it. And then, right as the Series A round closed in June 2021 with about $34 million from investors, The SEC alleges Saniger sold approximately three million of his own Nate shares.

GrantWhile the automation rate was effectively zero.

DerekWhile the automation rate was effectively zero. That's the moment this stops being optimistic lying about a product that might eventually work; that's a decision.

GrantThe math doesn't work, and at some point he had to have known it.

DerekHe knew-the dashboard told him-he just made sure nobody else could see it.

GrantSo how long can you actually hold that together before something breaks?

DerekOne reporter asking the right question. So that illusion held for almost three years. And here's what finally killed it.

GrantA journalist.

DerekOne story. The Information ran an investigation in June 2022 questioning Nate's AI claims. That's it. One article and the pending funding round evaporates immediately.

GrantWait, Wait, wait. A company built entirely on investors not asking hard questions collapsed the moment someone asked the hard question?

DerekBasically, according to the DOJ indictment, Nate failed to close at Series B after that report dropped. No new money, no runway. Company dissolved in January 2023 through a California assignment for the benefit of creditors.

GrantWhich is just the polite way of saying.

DerekEverybody loses. Investors took near total losses. Tens of millions gone.

GrantAnd Nate Sanger?

DerekHe already pocketed that three million from the Series A share sale we... we'll be covered. Walks away, company dissolved, he moves to Barcelona.

GrantHold on—he's in Spain?

DerekBarcelona, yeah; and here's where this gets genuinely strange. The DOJ and SEC filed parallel criminal and civil actions on April ninth, twenty twenty-five—securities fraud, wire fraud, up to twenty years on each count.

GrantOkay.

DerekAs of the This year the SEC still hasn't served him; they filed a letter motion in May, twenty twenty-five, asking for more time; the court gave him until July eighth to file a status report. According to Holland and Knight's analysis of the case, no responsive pleadings have been filed. The DOJ's only filing is the indictment itself.

GrantSo he's just there,

DerekThere!

Grantrunning a VC fund.

DerekManaging partner at Buttercore Partners. Still listed on the website.

GrantThe math doesn't work, and everyone knows it. You raise money on fake AI, the company dissolves, investors lose everything, you move to Spain, you get indicted, and then you just keep running a venture capital fund?

DerekThe indictment exists-Spain exists-those two facts are just sitting next to each other.

GrantAnd Global Investigations Review noted the DOJ and SEC are both intensifying their focus on AI washing enforcement. And this is supposedly the escalation?

DerekIt is the escalation-first criminal charges for AI washing under the current Administration; that matters, but the underlying question it raises is whether enforcement catches up before the next version of this pitch is already funded.

GrantAnd that's the pattern, right? Because Nate isn't alone.

DerekNot even close-and that's exactly what we need to get into. So here's where we pull all the way back. Kempelen's candle, Amazon's worker IDs, Saniger's standby engineers. Three centuries, same trick. The specifics change, the structure doesn't. And the enforcement is finally catching up, right? According to the SEC's own filings, Delphia and Global Predictions paid a combined $400,000 in fines back in March 2024, civil penalties. Then, in January, twenty twenty five, Presto Automation becomes the first public company charged.

GrantWow!

DerekAnd then, Nate, April, twenty twenty five, first criminal counts.

GrantCivil fines to felony indictments in about a year-that's not a slow escalation.

DerekNo, it's not, and Presto is worth pausing on because it's almost too on the nose: AI drive through voice assistant, right? Claimed it eliminated human order taking entirely. Similarly, the SEC found that the vast majority of those orders required human intervention-human agents in the Philippines and India processing the orders.

GrantA machine with a person inside.

DerekDeadpan. Literally: the mechanical Turk at a fast food drive through.

GrantAnd two thousand twenty three.

DerekSo why does it keep working? And I think the honest answer is structural, not about stupidity. You're in a hype cycle. Everyone wants to believe the magic is real, and software demos are genuinely hard to stress test from the outside.

GrantThe gap between "it works in demos" and "it works in production" is invisible unless you're actually trying to break it.

Speaker 3Mm-hmm.

DerekWhich nobody was doing. Darrow published a piece pointing out that AI washing follows the exact The exact same investor psychology as the dot com era, where adding the right word to your pitch invited both the money and the credulity.

GrantWe covered the same thing last month with TerraLuna. The math couldn't work, and the people in the room probably knew it, but they wanted it to work, so they decided to believe it.

DerekDesire to believe is load bearing in all of these.

GrantSo what would actually catch this? Like, what would have stopped Nate?

DerekHonestly, someone trying to break the product, or a journalist willing to look, which is what happened. The Information asked a real question, and the funding round died.

GrantThat's not reassuring as a system.

DerekNo, it's not. And Presto gets caught because an analyst finally read a disclosure that said plainly, humans are doing 70% of this.

GrantThe Global Investigations Review piece noted the DOJ, SEC, and FTC are now all coordinating on AI washing enforcement, so the response is real-whether it outpaces the next cycle is a different question.

DerekAnd that's the question, right? The con is old, the audience's appetite for autonomous machines is older, enforcement is new-which one has more momentum?

GrantI don't know the answer to that.

DerekNeither do I And that's the episode! The thing that keeps sticking with me is the candle ritual. Kempelen opens every cabinet door, lets you look inside, and the whole time a Chess Master is folded up in there! The transparency was the con!

GrantRight; and Senator ran the same move two hundred and fifty years later; raised over fifty million dollars on a promise the DOJ says delivered basically nothing on day one. Same structure, no wink.

DerekNo wink—that's the whole story actually. Amazon kept the wink—that's the line between a clever product and a fraud case.

GrantThe math didn't work and the dashboard that proved it was labeled a trade secret. Of course it was!" Grant is a financial analyst with a crisp, authoritative, neutral American delivery that sounds like a seasoned portfolio manager explaining a complex market shift. His voice has a smooth, polished texture and a steady energy baseline that projects confidence and intellectual rigor without being overly aggressive. He maintains a grounded demeanor that balances the skepticism of a veteran investor with the genuine curiosity of a tech enthusiast.

DerekIf this one got at you, tell someone who still trusts every pitch deck they read. Subscribe wherever you listen, drop us a review. It genuinely helps.

GrantThanks for being here. We'll see you next time.

More episodes

Keep listening

Other episodes of Vaporware, newest first.

All episodes of Vaporware

Sources

Where this came from

36 reports behind the episode. Every one of them opens where it was published.