Fighting $25 Million Deepfakes with Behavioral AI

Simon Carver

[authoritatively] Right now, inside the world's largest financial institutions, AI systems are analyzing TRILLIONS of dollars in movement patterns -- sanctions risks, ownership layers, payment instructions -- all in real time. But the criminals are using AI too. [warmly] Welcome to the show, everyone. I'm Simon Carver, here with Lachlan Reed and our guest, Jack Burns.

Jack Burns

We need to start by understanding how primitive the old system was. [calm] For decades, anti-money laundering -- AML -- ran on static rules. If a wire transfer crossed a certain dollar threshold, flag it. If money moved through a sanctioned geography, flag it. [sighs] The result was what investigators called the "false positive avalanche." In some banks, more than NINETY-FIVE percent of alerts turned out to be absolutely nothing.

Lachlan Reed

Ninety-five percent? [scoffs] [skeptical] You're telling me investigators were spending their entire careers chasing GHOSTS ninety-five times out of a hundred? That's... well, even a KANGAROO could trip over those odds. [chuckles] You're just clicking "ignore" all day.

Jack Burns

Exactly. It creates fatigue, and fatigue creates missed signals. [measured] But the new philosophy completely abandons those static rules. It relies on behavioral intelligence. It maps what we call the "behavioral topology of money."

Simon Carver

Topology. [thoughtfully] So instead of looking at one isolated wire transfer, it's looking at the ENTIRE social network of that money.

Jack Burns

Yes. [explaining] Timing relationships, counterparty networks, geographic flow sequences. The AI asks: "Does this behavior make sense inside the context of everything this entity normally does?" A legitimate multinational corporation sweeping liquidity between subsidiaries looks VERY different from a laundering syndicate layering transactions -- [deliberate] but only if you can see the WHOLE ecosystem.

Lachlan Reed

But you can't see the ecosystem if the data itself is rubbish. [excited] And this is where ISO 20022 comes in. Most consumers have never heard of it, but it's a massive infrastructure shift. Legacy payment formats were basically fragmented text strings. [matter-of-fact] ISO 20022 forces STRUCTURED data: ultimate debtor, ultimate creditor, explicit payment purpose, tax identifiers.

Jack Burns

That structure is the critical variable. [authoritative] AI is only as powerful as the quality of its data. With ISO 20022, AI can perform fuzzy matching against those ownership layers and tax records in milliseconds. [pauses] It finally has investigative VISION.

Simon Carver

[intrigued] Which brings us to the part that feels a bit like science fiction... Agentic AI.

Lachlan Reed

Right, because most people think AI just outputs text. You type a prompt, it spits out an answer. [matter-of-fact] But Agentic AI ACTUALLY executes workflows. We're talking about autonomous investigative participants.

Jack Burns

That distinction matters enormously. [quietly] Inside the modern investigative environment, these autonomous agents are now doing the mechanical labor. They retrieve ownership data, cross-reference sanctions databases, pull adverse media, and pre-score risk. [short pause] All in seconds.

Simon Carver

So an investigator sits down, and [realizing] instead of starting with a blank screen and having to pull all those records manually, they begin with a fully assembled intelligence package. The job goes from "data hunter" to "STRATEGIC ANALYST."

Jack Burns

Correct. But here is the friction point. The criminals are doing the EXACT same thing. [serious] We are seeing a dramatic spike in AI-generated phishing, synthetic identities, and deepfake executive audio.

Lachlan Reed

I read about a case recently where a finance worker wired out TWENTY-FIVE MILLION DOLLARS because he was on a video call with what he thought was his CFO and several colleagues. [disbelief] All of them were DEEPFAKES. Every single one.

Jack Burns

And static defense models CANNOT catch that. [urgently] If the credentials are valid, the system lets them in. That is why the defense systems themselves MUST become behavioral. Instead of checking for a malicious signature, the AI checks the human baseline. Does this CFO normally send urgent payment requests at this time of day? Do the typing cadences match? Are there vocabulary shifts?

Simon Carver

If the linguistic patterns are slightly altered, the system triggers human intervention. [warmly] Which brings us right back to the core premise of this show. Even with all this behavioral mapping and these autonomous agents, HUMANS remain the backstop.

Jack Burns

More essential than many executives realize. [contemplative] Because financial crime investigations are not purely mathematical. They require geopolitical awareness, legal interpretation, and human motive analysis. AI surfaces the pattern. [deliberate] Humans determine the MEANING.

Lachlan Reed

[thoughtfully] So AI might build the fortress, but humans still have to decide what the fortress actually PROTECTS.

Simon Carver

Exactly. [brightly] Thanks for joining us for this quick take on The Human Workforce Podcast. If you learned something today, make sure to LIKE, SUBSCRIBE, and SHARE the episode with someone who still thinks AI is just a glorified chatbot.

Lachlan Reed

And let us know what you think -- [challenging tone] will these systems make finance safer, or just create invisible digital power structures?

Simon Carver

[smiling] Until next time -- stay curious, stay HUMAN, and we'll see you in the next episode.