Fraud Fighters, Agent Wallets, and the AI Funding Fear Test

LaurenOkay, so get this. Somewhere right now, a piece of software is holding a wallet, and nobody's quite sure who gave it the keys.

DerekOh, man. Was it me? I once let a smart speaker order cat litter.

LaurenAnd we survived. But this is bigger than litter.

DerekWay bigger.

LaurenAnd honestly, my own startup days taught me the scariest moment isn't when money goes missing. It's when nobody can say who approved it, and everybody just stares at the ceiling.

DerekOkay, that is a very specific scar. Do we need a moment?

LaurenI'm Lauren, this is Derek, and this is Tech Insider Weekly, where today's question is simple: When AI touches money, who's watching it?

DerekBecause it's coming at money from every side. Some AI is out there hunting fraud. Some is being handed a budget to spend.

LaurenWait for it, though. There's also the money chasing all of it, and some people are nervous about that.

DerekNervous how?

LaurenOh, no, you'll hear it. I'm not spoiling the turn.

DerekTease.

LaurenAlways. Fine, Derek, where do we begin?

DerekSomewhere with stakes, a company whose whole job is catching people who lie to their bank.

LaurenOh, that's perfect. Let's start there. Okay, so the company is Casap, and Fortune reported it raised fifty-six million dollars to help banks deal with a surge in customer fraud.

DerekCustomer fraud, as in the customers are the problem.

LaurenSome of them, yes. Fortune says a social media-driven epidemic of dishonesty means card companies face far more first-party fraud.

DerekOkay, that term sounds jargony, so let me make it plain. Third-party fraud is a stranger stealing your card. First-party fraud is the person who actually owns the card doing something shady.

LaurenLike?

DerekBuying a pair of sneakers, wearing them to a wedding, then telling the bank you never ordered them. The bank has to prove you did.

LaurenHuh, wearing them to the wedding is a bold detail.

DerekI've seen worse.

LaurenAnd the fun part is where the dishonesty spreads: social media.

DerekOf course it does.

LaurenHere's my operator brain on this. When someone says they never ordered the sneakers, somebody at the bank has to gather the evidence and answer the dispute. That's real work, and it gets multiplied by every single case.

DerekAnd honest customers file disputes too, so the bank can't just assume everybody's lying. It has to tell the difference over and over at scale. That's the job a company like this is selling into.

LaurenNow, Casap's own release calls the round a fifty-six million dollar Series B, and it comes loaded with numbers. Up to fifty-one percent lower first-party fraud losses.

DerekUp to, two of my least favorite words.

LaurenSame. Then it says a ninety-six percent chargeback rate against a sixty percent industry average.

DerekWait, ninety-six against sixty? That's not a gap. That's a different sport.

LaurenRight? And disputes resolved in eight days instead of up to ninety.

DerekOkay, I'm impressed for about four seconds.

LaurenBecause?

DerekBecause who's grading this? The company is.

LaurenExactly. Those are Casap's claims in Casap's release, and nobody independent has checked them. So my question is always, but what actually happened? A win rate on whose disputes measured how?

DerekAnd the release doesn't say, at least not in what we've got.

LaurenNo. A big number with no method is a billboard.

DerekYou said billboard with a Series B.

LaurenI've sat through vendor demos where the slide was gorgeous and the footnote was missing. If I were a bank evaluating this, I'd have a short list. Which disputes are in that win rate, over what period, compared to what baseline, and who did the counting?

DerekAnd what does resolved in eight days actually mean? Resolved in the bank's favor or just closed? Those are very different outcomes.

LaurenRight. I'm not accusing anybody of anything. That's just what I'd ask before signing a contract.

DerekSpoken like someone who's been burned by a pretty dashboard.

LaurenThere's one more number, and it's the one that matters for our whole question. The release says seventy-three percent of dispute cases get resolved through agentic automation.

DerekAgentic meaning the AI does the work itself, not just suggests.

LaurenYes, roughly three out of every four cases.

DerekHold on. So the software handles most of it. Where's the person?

LaurenThe release says human oversight is retained. It doesn't spell out what that looks like.

DerekHmm. That's a pretty elastic phrase.

LaurenCould be a person approving every call. Could be someone glancing at a dashboard on Fridays.

DerekFair. Still, a bank where the AI does the grunt work and a human holds the final say is a lot more comfortable than the alternative.

LaurenAnd that's the thing about retained oversight. It's the phrase that sounds the most reassuring and tells you the least. If I were a buyer, I'd want it in writing what a human can actually override.

DerekRight, and how often they do, because a person who can technically stop the machine but never does is basically decoration.

LaurenWhich is a nice setup because our next company goes further. Casap keeps humans in the loop. Now a company that lets agents hold the wallet with approval.

DerekOh, you're gonna love this one.

LaurenOkay, so get this. Stripe just announced Treasury for Agents.

DerekTreasury, like the boring money plumbing?

LaurenThe boring money plumbing now with a chatbot at the controls. Stripe's own announcement says startups can have agents like ChatGPT and Claude check balances, pay invoices, move funds, and convert currencies.

DerekConvert currencies? Hold on.

LaurenYep. All of that runs through Stripe's MCP server.

DerekOkay, I get to nerd out for a second. MCP is basically a standard doorway. The agent knocks, the server says, "Here's what you're allowed to ask for," and the agent makes the request.

LaurenA doorway with a bouncer.

DerekSort of, but the sources only take me that far. I can tell you what it connects, not how Stripe built the locks.

LaurenWhy does a standard doorway matter, though? Say it like I'm not an engineer.

DerekFor the non-engineers in the audience, a standard means the assistant and the service can talk in a shared format instead of everything being hand-wired one connection at a time. That's the general idea. What Stripe actually exposes through it is the list in the announcement. Mm, I'll stop there.

LaurenFair. So what does a startup actually do with it?

DerekStripe's example is a company called Kubrio, and it's a one-founder company.

LaurenOne founder. So the finance department is a person and a chatbot.

DerekAnd the chatbot never asks for a raise.

LaurenBut seriously, think about what that means for a company of one. There's no finance team to notice a weird payment. The founder is the entire control system.

DerekAnd that cuts both ways. It's less paperwork for someone already doing everything, but it also means fewer sets of eyes on the money.

LaurenWait for it, though. There's a catch with the card side.

DerekRight. The release describes card creation as coming soon. That's not a button you can press today.

LaurenSoon, the most flexible unit of time in tech.

DerekAnd the guard rail is worse. Agent spending limits are described as arriving in the coming weeks.

LaurenHang on. Limits aren't live yet?

DerekPer the release, no. That's the feature that lets you say, "This agent can spend this much and not a dollar more."

LaurenLet me be the operator for a second. A limit is the thing you set before you trust anyone with money: new hire, contractor, anyone. You don't hand over the card and then discuss the cap.

DerekAnd an agent is a strange new kind of colleague. It doesn't get tired or embarrassed, so you can't count on it to hesitate the way a person might.

LaurenSo the safety rail is also on the roadmap. Love that for us.

DerekTo be fair, there's the other layer. FinTech Global covered the launch letting AI manage startup finances with human approval.

LaurenOkay, that's the part I care about. The human still signs off.

DerekYeah. The agent does the legwork, and a person hits approve.

LaurenWhich sounds great until you picture the approval queue at 5:00 PM on a Friday.

DerekOh, man. Approve, approve, approve. What was that one?

LaurenImagine handing an AI a company card. It's a very polite intern who never sleeps and has no idea what a latte budget is.

DerekAnd it'll pay the invoice in euros just to be thorough.

LaurenHonestly, though, the approval step is doing a lot of work in that pitch.

DerekIt is, and it only matters if somebody reads what they're approving.

LaurenThink about how a rubber stamp feels at the end of a long week. Nobody means to stop reading. It just happens one click at a time.

DerekSo the question for a founder isn't whether there's a human in the loop. It's whether that human has the time and the information to actually say no.

LaurenHmm. Okay, so that's the plumbing. Agents get wallets. Humans keep the approve button.

DerekRight. So the real money question is somewhere else, isn't it?

LaurenSwitching gears. Everybody's nervous about funding right now, so where are the checks actually going?

DerekOh, you're gonna love the answer.

LaurenOkay, so get this. Same morning, two headlines, and they don't seem to know about each other.

DerekGo on.

LaurenSiliconANGLE reported that an inference cloud startup called General Compute reeled in $100 million, and the headline itself says it happened despite broader AI funding fears.

DerekWait, the headline admits the fear?

LaurenRight in the title, like announcing a pool party with a weather warning.

DerekOkay. And for anyone who hasn't touched this stuff, inference cloud means what?

LaurenI'll stay in my lane. The headline says inference cloud, and that's all I'm claiming.

DerekFair. Honest. Rare.

LaurenAnd then get this: Silicon Angle also reported that Upwind, a cloud security startup, raised $400 million.

Derek400?

Lauren400 at a valuation of 3.8 billion, and the headline says revenue quadruples.

DerekQuadruples? Okay, that's a number with a reason attached, which I like better than a number floating in space.

LaurenBillboard callback. Nice.

DerekI contain multitudes.

LaurenOne thing I'd flag on Upwind, revenue quadruples is the headline's framing, and I don't have the starting number. Quadrupling from small is a different story than quadrupling from large.

DerekTrue. A great growth rate on a tiny base is still tiny, so I'll enjoy the number without pretending it tells me how big the business is.

LaurenTwo raises, one morning, and both stories carry the fear right alongside the money.

DerekSo what exactly is everyone afraid of?

LaurenThe Washington Post has a piece titled There's Not Enough Money in the World to Pay for the AI Boom.

DerekOh, man. Cheerful.

LaurenIt reports AI companies are borrowing hundreds of billions of dollars and selling new stock.

DerekHundreds of billions with a B.

LaurenAnd notice what that sentence is about. It's describing a pattern across AI companies, not saying any one of these startups is in trouble.

DerekWhich is different than a raise at a given startup. One is the whole sector's appetite, the other is a single company's check.

LaurenYep, and trying novel financing methods that it describes as potentially risky.

DerekNovel and risky, two words I never want next to my mortgage.

LaurenOr a power plant. The Post cites an estimated $10 trillion of AI-related spending in the coming years.

Derek10 trillion. I can't even picture that.

LaurenTry this. That's roughly 10,000 billions. So 100 billion is 1% of that.

DerekSo General Compute's 100 million is a rounding error in a rounding error.

LaurenHuh, pretty much, which is why I'm careful about what this proves.

DerekMeaning?

LaurenMeaning nothing about why these two got funded. The stories give me the raise and the worry. They don't give me a cause connecting them.

DerekRight. Money still lands on specific companies while the worry grows. Whether one is steering the other, nobody here says.

LaurenI keep thinking about the founder on the other side of this. You read both stories and have to decide what to plan around.

DerekThe raise or the worry?

LaurenNeither is a plan. Both are weather.

DerekWeather you can't control, so you control what you can, like what the money is for and how long it lasts. That's just my opinion, not something the stories say.

LaurenExactly. And there's a darker flavor of the worry.

DerekOf course there is.

LaurenA newsflash summary on KuCoin describes a Musk post predicting what it calls guppy-level death and consolidation among more than two hundred Neoclouds.

DerekGuppy level. I love that he went with the aquarium.

LaurenIt's a secondhand summary of a post, so treat it as a prediction. The same summary also cites an October Gartner forecast.

DerekWhich says?

LaurenIt says more than eighty percent of Neocloud providers could stop operating independently by twenty thirty.

DerekCould, not will.

LaurenRight. A forecast, not a body count.

DerekStill, that lands next to general compute in an awkward way. Neoclouds, inference clouds, same neighborhood.

LaurenThe summary itself calls it relevant to inference cloud startups like them, yes, but I won't pretend it says general compute is in danger.

DerekGood, because the hundred million just showed up in the bank.

LaurenCongratulations. You may now be consolidated.

DerekHappy raise day.

LaurenLet me say how I'm weighing all of this, because it's easy to overcorrect. If I only read the post piece, I'd think nobody can raise a dime. If I only read the raises, I'd think the worry is made up.

DerekAnd both are on the page the same day. Nobody's lying. They're just looking at different things, a couple of specific companies on one side and the overall spending picture on the other.

LaurenRight. Specific versus aggregate, and I'd rather sit in that discomfort than pick a side.

DerekSo what would change your mind either way?

LaurenDetails these stories don't give us, like who put the money in and on what terms. Until then, it's a raise and a worry side by side.

DerekWhich honestly is the most accurate headline we've got.

LaurenOkay, switching gears. Next, two quick separate items from EU startups.

DerekOne's an opinion piece on European AI funding.

LaurenAnd one is one small raise, very small compared to ten trillion. Okay, so the EU startup's opinion piece first. Its headline says scaling is the real test of Europe's AI boom.

DerekTest, like an exam?

LaurenLike an exam where not everyone gets the same questions. The piece says Europe's AI funding boom isn't benefiting every part of the market equally.

DerekOkay, but boom for whom? Give me a number.

LaurenRecorded Series A to C investment rose by approximately sixty-six percent in the first nine months. That's the one figure the excerpt gives me.

DerekHold on. The excerpt cuts off, right?

LaurenRight. So I can tell you the headline and that number. Who got squeezed out, I can't say.

DerekFair. A two-thirds jump in the middle round still tells you which stage is getting the attention.

LaurenMm, though that's your inference, not theirs.

DerekCaught me.

LaurenHonestly, the useful reading is the cautious one. A big percentage tells you something moved, not who benefited from it.

DerekAnd an opinion piece is an argument, not a census, so I'll hold the number loosely and the headline even looser.

LaurenSwitching gears. Totally separate item, same site. Magic Vision, an Italian deep tech startup, raised six hundred thousand euros.

DerekWhat do they build?

LaurenAI and computer vision to monitor and support older people and vulnerable individuals.

DerekOh, that's a real problem, not a pitch deck problem.

LaurenThe article title is a quote about someone's grandfather falling and staying on the floor for an entire day. That's the whole origin story in one sentence.

DerekOuch. Okay, that lands harder than any valuation.

LaurenIt's a human detail attached to a funding line, which is rare. Most raise stories read like spreadsheets.

DerekThough a good quote doesn't prove the product works. That's a separate question, and the headline doesn't answer it.

LaurenAnd notice it's small money, six hundred thousand euros. I'm not claiming it proves anything about the boom above.

DerekRight. Two items, no forced bridge.

LaurenBut if I'm giving founders one opinion, and it's only an opinion, it's this: Keep a human watching the money and know exactly which problem you're solving. That seems to travel better than a big headline.

DerekAgreed. A clear problem, a person in the loop, not a bad checklist for a week with robot wallets in it.

LaurenTrue. Okay, that's the week. Let's wrap it up. Okay, so the question we started with, who watches the money when AI's holding it?

DerekA person, ideally one who's awake.

LaurenLow bar, oui, yes. Whatever the round size, the stories that stuck were the ones with a human still in the loop.

DerekAnd a problem you can name in one sentence.

LaurenRight.

DerekHonestly, that's the thread through the whole episode. Fraud, agent wallets, big checks, all of it comes back to somebody being accountable.

LaurenAccountable and actually paying attention. Showing up in the approval queue isn't the same as reading it.

DerekSo if you're a founder listening, ask yourself who approves the spend before you ask who writes the check.

LaurenOh, and subscribe, please. It's free, and it keeps Derek from refreshing the download count.

DerekI do not do that. Okay, I do it a little.

LaurenLeave us a review too. Tell us which founders you want us to dig into or which topics we should poke at next.

DerekTag the show when you share an episode. We'll find it, and we'll be weirdly happy about it.

LaurenNew episodes land every Wednesday.

DerekLauren, I'll be here approving nothing without a human present.

LaurenBring snacks. See you Wednesday.