Nate: The AI That Was Just People

DerekPicture a phone. One tap, and your sneakers are bought. No cart, no card number, no checkout page. Magic.

GrantOkay, I want that app immediately.

DerekPlot twist. Prosecutors say the magic had a night shift. People typing it all in by hand.

GrantWait. People? Like actual humans?

DerekContract workers in the Philippines and Romania, according to the allegations.

GrantSo the AI was a guy named Whoever.

DerekHold on. We're getting ahead of ourselves. I'm Derek, this is Vaporware, and Grant is here to doubt everything I say.

GrantIt's my only job. And I'm Grant.

DerekQuick housekeeping, though. Everything we say here is an allegation. Albert Saniger, the founder of Nate, has pleaded not guilty.

GrantFair. So the claim is he told investors the app used proprietary AI to check out purchases on its own.

DerekRight. And prosecutors say the work was done by hand. So the whole case is the distance between what investors heard and what was allegedly happening behind the screen.

GrantThat's the gap. A pitch on one side, workers on the other, and a courtroom in between. Hmm. Okay, but there's gotta be a number somewhere. There's always a number.

DerekOh, there's a number. Wait for it. It shows up later, and it's a doozy. I'm not touching it yet.

GrantYou're a tease.

DerekOccupational hazard. But first, the setup, and it's a good one.

GrantSo what exactly was the pitch that got investors to write checks?

DerekOkay, so get this. The pitch fits on a sticker. Skip the checkout.

GrantThat's it? That's the whole thing?

DerekPretty much. The Justice Department's announcement of the indictment describes a single tap promise. You find something in any online store, you tap once, and the app does the rest.

GrantDefine the rest, because checkout is the annoying part. That's the part with the tiny form fields.

DerekExactly the annoying part. The DOJ gives a sneaker example. The app was supposed to pick your size, enter your billing details, enter your shipping details, and confirm the order, all of it using AI.

GrantSo I'm just a guy standing there with my thumb up.

DerekThumb up, wallet closed, and prosecutors allege the AI part is exactly where the story falls apart.

GrantHold on. Allegedly.

DerekAllegedly, yes. Saniger has pleaded not guilty. Now the money.

GrantRight. How much money are we talking?

DerekThe SEC says over forty-two million dollars.

GrantWhoa.

DerekIn its civil case against Saniger, the SEC alleges he raised that across the seed and Series A rounds, spring of twenty nineteen through December twenty twenty-two.

GrantWait, so that's two different cases, one civil from the SEC and one criminal from the Justice Department.

DerekTwo tracks. The SEC complaint is civil, the indictment is criminal, and both are about what the app could do.

GrantThat's almost four years. Four years of cashing checks on one sentence?

DerekWell, the SEC says the sentence was the problem. The alleged false claim was that the app completed purchases using AI without human involvement.

GrantWithout human involvement. So that's the line. Not mostly AI, not AI with a little help.

DerekZero humans in the loop, per the SEC's telling.

GrantOkay, but here's where I keep stump on. Forty-two million is a lot of people with a lot of lawyers and a lot of smart analysts. Didn't nobody check?

DerekHmm, honestly, that's the question the whole case hangs on.

GrantBecause in my world, if you sell a car, somebody pops the hood. Even a bad buyer kicks a tire.

DerekSure, but a demo is a hood that's welded shut. You see a clean screen, you see the order go through, and you feel like you've kicked something.

GrantAnd you're a person who wants it to be true.

DerekWait for it, because that's the trap. Everyone in that room wants the one-tap future. The app that spares you the form fields is a wonderful idea. Nobody walks in hoping to catch the founder.

GrantHuh. So the pitch does half the selling by being something you already wish existed.

DerekRight. Which brings us to the demos.

GrantRaising the money is one thing. What did those investors actually see on the screen?

DerekOh, you're gonna love this part. The pitch deck wasn't just we have AI. It had a whole vocabulary.

GrantOf course it did. What words?

DerekThe Berkeley Technology Law Journal commentary from April of twenty twenty-six summarizes the SEC complaint. It says Saniger told seed and Series A investors the app used artificial intelligence, machine learning, and neural networks.

GrantNeural networks. That's the buzzword triple crown.

DerekWait for it. Automation success rates of ninety-three to ninety-seven percent.

GrantNinety-seven. Not ninety-five, not most. That's a number with a decimal point confidence to it.

DerekAnd a range sounds like measurement. Somebody ran a dashboard. Somebody has a spreadsheet.

GrantAllegedly. And the spreadsheet is the thing in question. Ninety-three percent of what, though? Measured by whom? Over what period?

DerekThe commentary doesn't say how the figure was calculated, so I can't tell you. All we have is the claim.

GrantWhich is the point. A precise range with no visible method is a number you're trusting on faith.

DerekRight. Everything here is an allegation, and Saniger has pleaded not guilty. But now the demos, because numbers on a slide are one thing. Watching the app work is another.

GrantSeeing is believing. That's the whole reason a demo exists.

DerekHolland & Knight, a law firm, walked through the SEC's allegations about the demos. The SEC alleges Saniger had staff prioritize manual processing of test purchases during demonstrations.

GrantHold on. Standby?

DerekStandby.

GrantSo picture it. The investor taps the button, the founder is beaming, and somewhere an engineer is hovering over a keyboard like a pizza guy waiting on a ring.

DerekThat's the allegation, yes. The app looks like it's thinking. A person is the thinking.

GrantAnd from the investor's chair, what would you even see? A phone, a tap, an order that completes. Nothing on that screen tells you who finished it.

DerekWhich is why the next detail matters.

GrantOkay, but investors poke at things. Mine would. I'd say, "Let me try one myself."

DerekHere's where it gets sharper. Debevoise, another law firm, summarized the allegations. It says Saniger allegedly had staff prioritize manual processing of the investor's own test transactions.

GrantNo way.

DerekSo the investor's own test goes to the front of the line, like a restaurant that hears the critic is in the building. Exactly. And think about what that does. You test it yourself, you get the fast, correct result, and now you trust it more than any slide.

GrantBecause you checked. You did the due diligence, and the due diligence was the one thing rigged.

DerekAllegedly rigged, but that's the part that matters legally, I think. A fake demo is theater. A fake test of the investor's own hands-on check is a lot closer to steering the evidence.

GrantMm. In deal terms, the one control you run yourself turns out to be the one that was managed. Your confidence ends up resting on the exact thing that was handled for you.

DerekHold on to ninety-three to ninety-seven percent, because the number on the other side is about to land.

GrantIs it lower? Please say it's lower.

DerekJust a little. The number on the other side of that pitch, Fortune reported in April twenty twenty-five that Justice Department officials put the actual automation at effectively zero percent.

GrantZero. Like zero zero?

DerekZero. Not low, not disappointing. Effectively zero.

GrantHold on. Ninety-something on the pitch and zero on the inside. That's not a gap. That's a different planet.

DerekAnd remember, this is what officials say. Saniger has pleaded not guilty, so everything here stays an allegation.

GrantFair, but let me do the numbers anyway, because I can't let this go. Ninety-three to ninety-seven percent is a claim about how little human help the thing needs. Effectively zero is the opposite claim about the same product.

DerekSame product, opposite descriptions. That's the entire dispute in one line.

GrantOkay, so if the machine wasn't doing the checkouts, who was?

DerekPeople. A Berkeley Technology Law Journal commentary summarizing the SEC complaint says contract workers in the Philippines and elsewhere handled most orders. Most. Not a few edge cases where the AI tap out. Most. So picture the order coming in, and instead of a neural network, there's a person on the other end with the browser open filling in the form.

GrantDude, that's the sneaker thing again, just with a human typing it.

DerekRight. The tap on the phone was real. What happened after the tap is what the complaint disputes.

GrantAnd that detail matters for the pitch, right? Because AI that checks out purchases and contractors who check out purchases are different businesses.

DerekIn general, software gets cheaper per order as it scales, and hand labor mostly doesn't. I'm reasoning there, not quoting anyone, but that's the kind of difference investors would care about.

GrantOkay, and here's the other question I'd ask. A company has engineers, support staff, people in the hallway. Wouldn't somebody on the payroll notice the machine was a room full of workers?

DerekThat's where the next allegation comes in.

GrantUh-oh.

DerekA law firm summary from Debevoise says Saniger allegedly told engineers to hide the state of AI development from other employees.

GrantWait, wait. Hide it from his own staff?

DerekThat's the allegation. Keep the engineers who know quiet, and keep everyone else guessing.

GrantSo the secret ingredient was a need-to-know policy.

DerekBasically a magic show where the stagehands sign non-disclosure forms.

GrantHmm. But that cuts both ways, right? If you have to hide it internally, you know what the truth is.

DerekThat's exactly how prosecutors would frame it. The defense will have its own telling, and he's pleaded not guilty.

GrantSure. And the people who weren't told, what were they supposed to think?

DerekPresumably that the AI was coming along, which if you're an employee, is a pretty comfortable thing to believe.

GrantMm. Everybody wants the story to be true.

DerekAnd the story only has to hold as long as the money keeps arriving.

GrantWhich it did for years, going by the SEC's window of twenty nineteen to twenty twenty-two. So something had to give.

DerekSomething did, and it started with a reporter. Then how did it come apart? June twenty twenty-two, a news article lands, and it questions whether Nate's AI is doing what the company says it does.

GrantWh-which outlet?

DerekThe Information. Holland & Knight's writeup of the case flags that report, and the Debevoise summary says it's said to have hurt later fundraising.

GrantSaid to. So that's the allegations side, not a court finding.

DerekRight. Everything here is the way the law firms describe the government's case. Saniger has pleaded not guilty.

GrantFine. But think about the timing. The company had been raising money for years on the automation story.

DerekMm-hmm. For years, yes.

GrantThen one article shows up, and the next check is reportedly harder to get. If I were charting this, that's the inflection point. Years of raising, then one report.

DerekAnd the dates are worth a look. The SEC's window for the raise runs through December twenty twenty-two, so June twenty twenty-two sits inside it.

GrantSo the article didn't end the raising on the spot.

DerekI can't say that from what I have. The Debevoise summary says later fundraising was hurt, and it doesn't give amounts or dates beyond that. But the article does read as a hinge.

GrantFair. And then what? Does the company just fade?

DerekSame summary says Nate stopped operating in twenty twenty-three.

GrantStopped. Like, lights off.

DerekLights off. And investors were left with large losses.

GrantHow large? Because we've been throwing around a pretty big raise.

DerekThe summary says large and doesn't put a figure on the losses, so I'm not going to invent one.

GrantHuh. That's honest at least. And it matters for the legal angle. Startups fail all the time, and investors know that going in. What's alleged here is that the claim they were funding wasn't true.

DerekRight. Failure alone isn't the allegation. It's what investors were told. And that's the sad part too. Somebody put real money into a one-tap future, and what they got was a shutdown.

GrantHmm. Believing the demo turned out to be expensive.

DerekRight. No stock photo of a happy ending here.

GrantSo the sources give us a June twenty twenty-two article, a shutdown in twenty twenty-three, and now a criminal case.

DerekPlot twist though, the company dying isn't the end for the founder. It's the start of a different problem.

GrantOh, I'm guessing that's the part with the courtroom.

DerekNow what does a founder in this position do next? So the plea. Saniger pled not guilty to the fraud charges over his claims about Nate's proprietary AI system.

GrantOf course he did. Everybody's day in court starts there.

DerekRight. And here's the detail I love. MLex reports he was released on a two hundred and fifty thousand dollar bond that lets him live in Spain.

GrantSpain? Wait, wait. The defendant in an American fraud case gets to stay in Europe.

DerekWith travel restrictions and other financial conditions. MLex doesn't spell out the rest, so I'm not going to invent it.

GrantFair. Still, a quarter million bond against the over forty-two million the SEC says was raised? That's well under one percent. A parking deposit.

DerekA very expensive parking deposit.

GrantOkay, so what is the actual fight? Because the demo trick isn't new. People have faked tech since forever.

DerekSure, but faking it in a pitch deck and getting charged for it are different things. The case is described to us in the brief behind this episode as the first criminal test of whether a human-powered AI claim counts as securities fraud.

GrantFirst criminal test, huh? And that's the brief talking, not some outside ruling.

DerekRight. Treat it as how the case is framed for us, the first time a prosecutor has to convince a court that it's a crime.

GrantSo what does the court actually have to decide?

DerekWithout playing lawyer, the allegation is that investors were told the app completed purchases using AI without human involvement and that humans did the work. The open question is whether that counts as securities fraud.

GrantAnd the downside for him?

DerekThat same brief says he faces up to twenty years. That's the maximum, and I can't point you to a second source for that.

GrantTwenty years? Okay, that's a lot for a shopping app.

DerekAnd the brief also says a November court date is set. Again, same caveat. That's all we have on the calendar.

GrantSo that's the next real moment, holding it loosely.

DerekSmart. And it isn't only a first for the courts. Holland and Knight, in a July twenty twenty-five analysis, calls it the first AI washing enforcement under the Trump administration.

GrantFirst under this administration. That's a narrow label, so I'll leave it at what the firm says.

DerekFair. But think about what a case like this means for any founder who's ever stuck a little AI sticker on a product that's mostly interns. Every big AI claim in a pitch deck now has a reader who might ask for proof.

GrantI'd bet a few of them are reading this docket right now, quietly, with the door closed.

DerekAnd all of it stays allegation until a court says otherwise. He's pleaded not guilty.

GrantRight. So, Derek, zoom out. Is Nate a one-off or part of a pattern?

DerekGood question. Let's look at the pattern. Start with what we've already covered on this show. Joonko, the AI hiring startup, and Beyond Alpha, the pre-IPO and AI fund pitch. Different companies, different mechanics, but the same question about whether the AI label was carrying more weight than whatever was behind it.

GrantAnd the Nate version is the oldest pitcher in the book, the chess-playing Mechanical Turk.

DerekThe cabinet with a person folded up inside. Plot twist, it's a Philippines and Romania version.

GrantWith better Wi-Fi.

DerekAllegedly. Saniger has pleaded not guilty, and nobody's proven the cabinet had anybody in it.

GrantFair. But what's the difference between those earlier two and this one? Because the mechanics really were different.

DerekHere, the thing being sold was the machine itself, the claim that software did the work. A Berkeley Technology Law Journal commentary from April twenty twenty-six files this case under AI washing in venture fundraising, and its title asks whether the AI was in the code or just in the valuation.

GrantThat's a rough title for a founder to read.

DerekIt's also a wider debate than this one case, which is why Nate reads as a data point, not a weird accident.

GrantA data point with a courtroom attached. But I don't want anyone walking away thinking all AI is a cabinet with a person in it.

DerekNo. Real AI products exist, and they get funded for good reasons. The pattern is narrower than that, though. The label travels faster than the proof, and a slick demo is where the two come apart.

GrantSo the question for the next founder with a slick demo is how many humans are in the cabinet?

DerekAnd whether a court will call that answer fraud. There's a November court date.

GrantOkay, then I think that's our cue.

DerekIt is. Time to sign off. So one last thing before we go. The whole story comes down to a single question for a courtroom. Does a person behind the curtain count a securities fraud?

GrantAnd nobody knows yet, which is wild given how old the trick is.

DerekAncient. Only the logo changes.

GrantThe logo and the valuation.

DerekHa. Fair. And remember, everything we said about Saniger is an allegation. He's pleaded not guilty, and that's where it stands.

GrantOne takeaway before the ask. An AI label is cheap to print, so ask what's standing behind it. And here's the ask, a practical one. Think of the friend who forwards you every pitch deck with three rocket emojis.

DerekOh, I know exactly who that is. I have two of them.

GrantTell that friend about this episode, then ask them one thing. Who's actually doing the work?

DerekAnd if you like the show, subscribe and leave us a review. It helps other people find us before they wire anything.

GrantNext time, another story about AI and what's actually behind it.

DerekPeek behind the curtain first. I'm Derek.

GrantI'm Grant.

DerekThanks for listening.