[warmly] Welcome to the show. Picture this: your phone lights up, you answer a video call, and your boss is staring back at you from a familiar office, same voice, same face, same little head tilt they always do before asking for something urgent... and that person does not exist. This quick take is called “Synthetic HUMINT: When the Human Is No Longer Human.” It’s one of those episodes that may change how you look at every screen you use. If you find this useful, like, share, and subscribe to The Human Workforce Podcast. I’m Simon Carver, here with Lachlan Reed and Jack Burns. Jack, good to have you back. [calm] Good to be here, Simon. [curious] Yeah, this one’s a bit of a brain-bender. We’ve talked about AI risk before, but this is different. This is the moment where a face on a screen stops being proof of anything. And mate, that’s a wobbly bit of ground to stand on. For a long time, human intelligence—HUMINT—meant something very physical. A handshake. Eye contact. The tiny pauses in somebody’s voice. We treated those signals like reality itself. So what exactly breaks here? [matter-of-fact] The category of the problem changes. Traditionally, you were asking, “Is this person telling me the truth?” Synthetic HUMINT forces a prior question: “Is this person real at all?” That is not simply better deception. It is simulation. And once simulation becomes real-time—video, voice, live response—you can no longer rely on human perception as your primary control. “Is this person real at all?”—that’s the line that sticks. I’m never gonna hear a clean video call the same way again. Because we’re not talking about the old dodgy deepfake clips where the mouth looked like it was chewing a sock. We’re talking real-time synthetic identity. A fully interactive person who can answer questions, react, stall for time, even chuck in little bits of personality. [questioning tone] And “real-time” is the key token there, right? Not a pre-recorded scam clip. Not a single fake voice note. A conversation. Exactly. Real-time means the system adapts while you are engaging with it. And agentic AI makes that adaptation more dangerous. Generative AI produces an output—a message, an image, a voice sample. Agentic AI maintains behavior over time. It can preserve context, track relationships, respond across channels, and continue the persona without a human operator manually puppeteering every exchange. [responds quickly] So the puppet master can knock off for lunch and the fake person keeps working. That’s the bit that feels properly cooked. In the old spy version, building a false identity—a legend—took ages. Documents, backstory, habits, all that. Now the system can spin up the lot in seconds and keep it alive. And that changes the emotional texture too. I mean, a lie used to have strain in it. Somebody had to remember what they said yesterday. They had to keep the story straight. But if the system is tracking everything, the consistency might actually be better than a tired human’s. [skeptical] In some cases, yes. Humans leak. We contradict ourselves. We forget dates, phrasing, small details. An engineered persona can be unnervingly stable. That stability is persuasive because our brains read consistency as trustworthiness. The problem is that consistency is no longer evidence of authenticity. Consistency used to be the badge. Now it might be the costume. [short pause] That’s rough. It’s like finding out the trail signs in the bush were painted by the bloke trying to get you lost. Let me try to say it back—slightly wrong so you can fix me. Synthetic HUMINT is basically when AI pretends to be a person convincingly enough that our normal social instincts—voice, face, timing, familiarity—start working against us? Almost. The missing piece is persistence. It is not merely “pretending convincingly.” It is maintaining a relationship credibly over time. That is why this matters beyond fraud. If I can create a synthetic human who exists across calls, messages, social feeds, shared contacts, and work interactions, I am not faking a moment. I am manufacturing a social reality. [softly] Manufacturing a social reality. Yeah... even a kangaroo could trip over that one. Because once you can see them, hear them, and video-chat with them, most people think the case is closed. Human confirmed. Done. But it isn’t done anymore. That’s the shock, really. Seeing is no longer believing. Hearing is no longer believing. Even a live call—something we’d usually treat as the gold standard—doesn’t guarantee the person exists in the ordinary sense at all. [calm] And this scales in ways people underestimate. A single synthetic identity is useful. A multi-agent network is strategically powerful. Instead of one fake executive, you create the executive, the colleague who vouches for them, the recruiter who mentions them, the mutual connection who confirms they are “great to work with,” and the vendor contact who has “been on calls” with them before. Different personas, same coordinated narrative. Social proof becomes synthetic. The phrase “multi-agent network” sounds technical, but the human meaning is simple: the room agrees with itself. Even if the room is fake. That’s what makes this chilling. You’re not being pressured by one liar—you’re being surrounded by a cast. [skeptical] And that cast can turn up anywhere work happens. Hiring’s the obvious one. Candidate looks great, sounds sharp, references check out—except the references might be part of the same synthetic network. Then internal comms. “Jump on this urgent call.” “Approve this transfer.” “Share that file.” Once the badge and the face stop meaning what they used to, every workflow gets a bit slippery. Precisely. Hiring, executive communication, vendor onboarding, customer engagement, even routine approvals. The vulnerability is not confined to intelligence agencies or geopolitics. It enters the enterprise through ordinary trust assumptions. And when organizations make decisions based on entities that do not exist, that is a governance failure before it is a technical failure. “Governance failure” is the right phrase. Because if a company responds by telling employees, “Just be more careful on Zoom,” that’s not a control. That’s basically handing someone a paper umbrella in a storm. [chuckles] A Bunnings umbrella in a cyclone, yeah. So what are the actual controls? Because “trust no one” sounds dramatic, but people still have to get work done. The shift is from trust by perception to trust by verification. You do not trust what appears human. You trust what can be proven. That means verifiable origin for communications—cryptographic signatures, authenticated channels, identity systems that are mathematically validated rather than visually assumed. Simon called it once “HTTPS for humans.” That is a useful analogy. HTTPS for humans—that one sticks. The padlock icon, but for a person. Not perfect, but much better than “I recognize the face.” And then there’s the other layer you’ve talked about before: metadata. The digital exhaust. Yes. Timing anomalies, rendering artifacts, behavioral irregularities, channel inconsistencies. Small traces that do not fit the claimed identity. Ironically, AI may be necessary to detect AI at scale. Human review alone will not be sufficient when the volume is high and the simulation quality continues to improve. So we end up in a weird spot where the machine that muddies the water is also helping test the water. Fair enough. But I keep coming back to zero-trust thinking. That phrase makes people tense up, because it sounds cold. Like you’re turning work into an airport security queue. [reflective] It is uncomfortable, but discomfort is not the same as error. Zero-trust, in this context, does not mean paranoia. It means no interaction inherits trust automatically from appearance alone. Each high-stakes interaction must earn trust through validation. That is more disciplined than cynical. And maybe more humane, strangely enough. Because if you design trust into the system, you’re not asking employees to become detectives every hour of the day. You’re giving them rails. They can still be warm, fast, collaborative—just not naive. That’s a good distinction. We’re not trying to turn everyone into a twitchy little lie detector. We’re saying instinct can’t carry the whole load anymore. In a world of synthetic people, instinct is a starting point. Design is the safety net. And the strategic danger, to be clear, is broader than a single successful deception. The deeper objective is erosion. Not merely to make you believe a lie, but to make you doubt the truth. Once verification feels exhausting, people stop trying. That is when manipulation becomes efficient. [softly] Reality apathy. That’s the thing I don’t want us to normalize. The shrug. The “who knows what’s real anymore?” posture. Because once a workforce falls into that, the system doesn’t just get less secure—it gets less human. And the winners, if there are winners here, will be the organisations that move early. Not because they’re scared of AI, but because they respect what trust is worth. Trust by instinct got us this far. Trust by design gets us through the next bit. [matter-of-fact] In a world where almost anything can be generated, only what can be verified will reliably matter. [warmly] That’s where we’ll leave it. If this episode gave you something to think about, like, share, and subscribe to The Human Workforce Podcast. Send it to someone working in hiring, security, operations, or leadership—they’re going to run into this sooner than they think. Lachlan, Jack, thanks as always Good one. Catch you next time, folks. Until next time.
Audio Coursesby Jellypod