AI will do wonderful things. Equity markets are likely to soar. But something else is highly likely with AI adoption: we’re all going to lose a lot of money. Through accident and by fraud, in the trillions of dollars.
This will show up in three ways: increasing stakes for agent malfunction, compounding growth in fraud complexity, and the continued globalization and fragmentation of money movement.
The stakes for agent malfunction will increase
Agents will malfunction. Perhaps not forever, but for a long time and in unpredictable ways. Whether due to an inherent tendency for hallucination, insufficient context, or outright error due to their inherent complexity, LLMs will continue to do things wrong. When it’s buying the wrong movie ticket, that’s one thing. When it’s sending $10K to the wrong address with finality of payment, that’s a whole other ballgame. The more we ask of Muse and Instinct, the higher the stakes for what they can eventually cost us.
Fraud will become even more convincing
Agents will revolutionize the business of fraud. It depends on the scam, of course, but in large part this is a labor-constrained operation. With AI, labor is now effectively infinite. It can speak any language without accent, effortlessly localize, and highly customize to the individual based on each of our digital footprints.
Globalization and fragmentation make everything harder
Real-time money movement currently sweeping the world makes the problem much harder. Pick your payment network of the future: RTP, Pix, UPI, SEPA Instant, stablecoins. These are all push networks. Once money is moved, it’s gone. No givesy’s backsey’s.
But wait, it's actually worse than you think. Fraud is already a terrifyingly uncontained problem. The internet made this a global business, meaning there is an incredible fragmentation of commerce and payment platforms through which transactions are conducted. And law enforcement is overwhelmingly local. I’m sure the FBI works very hard to protect us from internet evil-doers, but I’m also sure it’s very difficult for them to raid call centers in Southeast Asia. With AI, those call centers become open-source models running on laptops.
The biggest tell that fraud is out of control is that no one knows how big of a problem it actually is. The FBI recorded $21B of loss due to cybercrime in 2025 and has observed a 4-year CAGR of 32%. It also freely admits that most victims never file, making this not even remotely a real number. Interpol's (mostly consumer) survey measured $1T in losses worldwide in 2023. On a roughly apples-to-apples basis, they see year-over-year growth in the US and EU of well over 40% in essentially a pre-AI era. Yikes.
Bad actors need better defense
We need to do something. We need to lock down our critical systems. We need to secure the places that house and transmit our money and our data. Those places also need to communicate with each other directly about bad actors, and they need to do this globally and in real time.
In an age of agentic abundance, we see so many different opportunities to construct new means of protection and defense. While we’re just getting started, Foundation has been busy building our portfolio here. We’re proud to be early backers of companies like PayOS, Aegis AI, Skyflow, IPID, Stacklet and AI or Not.