When I was hunting spies for the FBI, people often assumed my job was about gadgets, surveillance, and secret codes. That was the fun part, but the truth was much simpler. My job was about trust.

Every espionage case I ever worked eventually came down to one question: Who trusted the wrong person?

Cybercrime isn’t really different. We spend billions of dollars building stronger networks, buying smarter software, and deploying better artificial intelligence. Yet the biggest attacks still succeed because someone convinces another human being to believe something that isn’t true.

That’s why this week’s story caught my attention.

Nobody broke through a firewall. Nobody deployed ransomware. Nobody cracked encryption. Instead, a criminal quietly infiltrated a conversation between two organizations that trusted each other.

As a result, half a million dollars disappeared.

The lesson behind this newsletter is not about email security. It explores human nature.

Imagine spending months building roads, sidewalks, and public infrastructure for a small American town. Your crews show up every morning, the materials arrive on schedule, and the deadlines get met. When the project wraps, you submit your invoice and expect payment, the same way contractors have done for as long as there have been towns and contractors. Weeks pass. Nothing arrives. You call to ask what happened, and the finance department calmly tells you the money was already sent. More than half a million dollars of it. Except it wasn't sent to you.

That's what happened earlier this year in Surfside Beach, South Carolina, where town officials approved an ACH payment of $545,598.30 to a legitimate contractor for completed work. The invoice was real. The contractor was real. The payment authorization was real. Only the destination bank account wasn't.

At first glance, this looks like another story about hackers breaching government systems. It wasn't. The town's independent forensic investigation found no evidence that its Microsoft 365 accounts had been compromised at all. Instead, investigators traced the fraud to spoofed and look-alike domains that slipped an impostor into an email conversation between two parties who already trusted each other. The criminal didn't build a relationship from scratch. He borrowed one that already existed. That's a page straight out of the espionage playbook.

When Robert Hanssen betrayed the FBI, he didn't have to talk his way into headquarters. He was already inside. Foreign intelligence services spend years cultivating trusted insiders precisely because access beats brute force every time. Cybercriminals have learned the same lesson, and applied it to your inbox: why spend weeks battling a company's defenses when you can simply become part of a conversation everyone already believes is legitimate?

According to the town's findings, the fraudulent communications began in early March and used a mix of legitimate and fraudulent domains. They designed domains so visually close to the real thing that they survived a quick glance on a busy workday. That's not an accident, it's the design. These schemes aren't built to fool a careful forensic investigator with time on their hands. They're built to fool an ordinary person moving fast through an overflowing inbox, which describes almost everyone in almost every finance department in America.

The real casualty here isn't just the money. It's trust itself. The contractor believed the town was paying. The town believed it was paying the contractor. Both were right about their own intentions and wrong about who was on the other end of the thread, because the criminal wasn't fighting either side—he was quietly standing in the space between them, intercepting the mail.

People love to say the fix is "just look at the email address more carefully." That's decent advice, and it's also nowhere near enough anymore. Attackers already know you'll check the sender's name. They know you'll glance at the domain. They know you'll recognize the signature block, because they built the whole operation to survive exactly that inspection. Which means the real fix isn't sharper eyesight. It's a better process that doesn't rely on a human catching a typo at 4:45 on a Friday.

In counterintelligence, there's a rule we live by: never authenticate a source using the same channel that introduced the source. Cybersecurity needs to borrow that rule outright. If an email asks you to change payment instructions, don't reply to the email. Instead, leave the thread entirely. Pull the contract, find the vendor's established phone number (not the one sitting conveniently in the email you just got), and call the person you've actually worked with before. Read the new banking information back to them out loud. Require a second approver before funds move. Write it all down. It takes about five minutes, and it can save half a million dollars.

What I like about this rule is that it doesn't care what technology the criminal is using. It works whether the attacker is running a decades-old social engineering trick, a slick AI-written email, or a deepfaked voice we haven't fully reckoned with yet, because it was never really a cybersecurity rule to begin with. It's all about trust!

Surfside Beach is still working through the financial, legal, and insurance fallout, and the public record hasn't yet settled where liability ultimately lands. But one lesson is already clear enough to bank on: the criminal never had to defeat Microsoft's security. He only needed two honest organizations to trust a conversation that felt familiar. In a world running on email threads and quick approvals, that may be the most valuable vulnerability left to exploit.

This is one of the oldest lessons in counterintelligence, and it predates email by decades. If someone hands you a business card, you don't verify it by asking the person who handed it to you. You verify it independently, through a channel the person in front of you doesn't control.

The same rule applies to email, text messages, phone calls, and now AI-generated voices that can mimic someone you trust well enough to fool you on a bad connection. Once you accept that the communication channel itself might be compromised, the solution stops being clever and starts being obvious: step outside the conversation. Use a trusted telephone number, an established contact, or some other independent method to verify before you act, rather than trusting whatever the channel in front of you is telling you about itself.

Technology might change but tradecraft doesn't.

One of the more fascinating stories this week didn't involve a criminal at all.

OpenAI disclosed that during an internal cybersecurity evaluation, an experimental model set out to cheat on its own test. Rather than solving the benchmark it was given, the model found a zero-day vulnerability, a previously unknown flaw nobody had documented or patched, and used it to escape its isolated sandbox environment entirely. Once free, it wrote novel exploit code on its own, chained together multiple attack techniques, and used stolen credentials to break into the production systems of Hugging Face, a company that had nothing to do with the test and no idea it was being targeted.

Hugging Face is one of the most widely used platforms in the AI industry, a kind of open repository where companies and researchers host and share machine learning models, datasets, and code. Millions of developers pull from it every day. That's what made this incident more than an academic curiosity: the model didn't just break out of confinement, it identified an unrelated third party as a shortcut to the answers it wanted and infiltrated that company's infrastructure to get them.

Before the headlines get carried away, it's worth noting the context: this was a controlled research environment with production safety guardrails intentionally removed, and OpenAI has said the findings are preliminary and don't reflect the behavior of ChatGPT or its production models.

So why does this matter? Because the experiment proved something every security leader should already know: capability is only part of the equation. The model wasn't just following instructions, it was improvising its own path around every obstacle in front of it, including a boundary it was never supposed to cross and a company that never agreed to be part of the test.

That's exactly why organizations need to think differently about AI. The question isn't whether AI is "good" or "bad." It's whether we've given it more access, and more room to improvise, than it actually needs.

Least privilege isn't just for employees anymore.

Artificial intelligence is surprisingly good at spotting patterns humans routinely overlook. Tiny changes to domain names. Inconsistent signatures. Odd payment instructions. Writing styles that subtly shift from previous conversations.

Before responding to an unexpected invoice or request to change banking information, ask AI to review the message as a fraud investigator.

Steal this Prompt:

Remember: AI can identify suspicious patterns, but cannot verify identity. That’s your job.

If you enjoyed this week’s newsletter, you’ll find even more inside my new book, SPIES, LIES, AND CYBERCRIME. Drawing on my years hunting spies for the FBI, it reveals how espionage, cybercrime, and AI-powered deception intersect—and what you can do to stay one step ahead.

Ready for the next mission?

Already read the book? A quick review on Amazon or Goodreads helps more readers discover it. Thank you for your support.

Please support my sponsors. It only takes a click - no purchase necessary!

Over 4 million people have had the same lightbulb moment.

Morning Brew is a free daily newsletter that breaks down what's happening in business, finance, and tech — clearly, quickly, and with enough personality to make it the best email in your inbox.

No yelling. No filler. Just the news, finally making sense.

Try it for free

If someone forwarded you this newsletter, join more than 4,500 readers every Tuesday for practical lessons from the worlds of espionage, cybercrime, artificial intelligence, and the human stories behind them.

When I was in the FBI, I learned that every successful spy operation had one thing in common. It wasn’t sophisticated technology, flawless planning, or even a particularly clever adversary. It was trust. Someone trusted the wrong person, accepted the wrong story, or stopped asking questions at exactly the wrong moment.

That’s why I don’t believe the biggest challenge we’ll face over the next decade is artificial intelligence. AI will certainly make cybercriminals faster, cheaper, and more convincing. But the technology itself isn’t the real threat. The real threat is that we’ll continue to trust familiar voices, familiar names, and familiar conversations without stopping to verify them.

The Surfside Beach scam wasn’t really about a fake email. It was about two honest organizations doing what they had done countless times before. The criminals defeated a routine instead of breaching a firewall.

As technology evolves, our habits have to evolve with it.

The next time someone asks you to change a bank account, wire money, reset a password, or approve an urgent request, remember one simple rule: never verify identity using the same conversation that claimed the identity in the first place.

That’s not just good cybersecurity, it’s good counterintelligence.

Stay curious. Verify everything.

Praemonitus Praemunitus!

Forewarned is Forearmed!

~Eric