AI vs Corruption: The Invisible Handshake

Simon Carver

[warmly] Welcome to the show. Once upon a time, bribery lived in envelopes, hotel bars, shell invoices, and the sort of back room nobody put on the org chart. Now it can hide inside a perfectly normal payment file moving through global systems at machine speed. Tonight on The Human Workforce, this is “The Invisible Handshake: AI, Corruption, and the New Financial Battlefield.” If you like what we do, like, share, and subscribe -- it really helps people find the show. I’m Simon Carver, with Lachlan Reed and our guest, Jack Burns.

Simon Carver

[curious] Jack, I want to start with a line that hit me hard from our prep: most companies think compliance is protection, but a lot of compliance programs are basically theater. Not fake exactly... just performative. Explain that.

Jack Burns

[calm] Because paperwork is not perception. A company can have policies, attestations, signed vendor forms, annual training, escalation thresholds -- all the artifacts of seriousness. And still miss the actual pattern. Modern corruption is rarely stupid. It is designed to appear legitimate. A consulting invoice, a local intermediary, a licensing payment, a vendor with clean paperwork. The deception is no longer in the surface of the transaction. Corruption today isn’t hiding in the transaction. It’s hiding in the context.

Lachlan Reed

[questioning tone] That word -- context -- that’s the whole meat pie, isn’t it? Because if you’re a bank or a big multinational pushing, I mean, billions of transactions across payment rails, no human team can eyeball every “consulting services” invoice and go, “Hang on, this one smells funny.” That’s like trying to spot one dodgy bolt in a shipping container full of bikes. You’ll go cross-eyed.

Jack Burns

[matter-of-fact] Exactly. And the sophisticated actor understands the thresholds. Keep the amount below escalation. Use an approved vendor. Ensure the contract is signed. Avoid sanctions lists. Use ordinary language. Nothing triggers the old rules because the old rules were built for obvious violations. Corruption adapted.

Simon Carver

[reflective] So the problem isn’t that the payment looks dramatic. The problem is that it looks boring. And boring scales. That’s the unsettling part for me. Humans end up reduced to checkbox operators while dashboards give executives the comforting illusion that somebody, somewhere, understands what’s happening.

Lachlan Reed

[lightly] Yeah, the old “green light on the screen means she’ll be right” routine. [short pause] But if the vendor was formed three weeks ago, or the owner’s cousin is tied to a port authority official, or the payments all bunch up right before a license renewal -- that’s not one document, that’s a web. Humans are decent at stories. We’re rubbish at invisible webs across five systems.

Jack Burns

[skeptical] Which is why this is not fundamentally a paperwork problem. It is an intelligence problem. Intelligence asks different questions. Not “Is the form complete?” but “What relationship does this payment sit inside? What timing pattern surrounds it? Which ownership structures overlap? Which actors repeat across jurisdictions?” If you ask bureaucratic questions, you receive bureaucratic answers.

Simon Carver

[softly] And if the real crime is relational, a form was never going to catch it. That’s the handshake nobody sees.

Lachlan Reed

[energized] Alright, let’s make this concrete. A payment comes in: “consulting services.” Amount is under the threshold. Vendor is approved. Paperwork’s tidy. No sanctions hit. Human reviewer sees a clean lane. But an agentic AI system -- and yeah, even saying “agentic” makes me feel like a kangaroo on roller skates -- it can look sideways, not just down. It sees the vendor entity was formed three weeks earlier. It sees shared ownership metadata. It spots family ties to a port authority official. Then it clocks similar payments in other regions landing around licensing renewals. That’s not one odd invoice anymore. That’s choreography.

Simon Carver

[leaning in] “Three weeks earlier” is the bit that sticks for me. Because a person reviewing one invoice won’t naturally connect formation date, family tie, and timing around a port license. They’re not stupid -- they’re just trapped inside one window.

Jack Burns

[calm] Correct. Human review is bounded by attention and system design. If the data lives in silos -- payments here, vendor onboarding there, beneficial ownership elsewhere, external watch data somewhere else again -- the investigator is effectively blindfolded and asked to detect a constellation. AI, used properly, can map relationships across those silos. It can monitor beneficial ownership structures, cross-reference external datasets, analyze behavioral anomalies, and identify linguistic intent, including coded or evasive language.

Lachlan Reed

[curious] The coded language bit is fascinating. Not because the machine reads minds -- it doesn’t -- but because it can notice when language shifts, right? Like when normal procurement chatter suddenly turns weirdly indirect or sentimental or overly vague around the exact time money moves.

Jack Burns

[measured] Yes. Not clairvoyance -- pattern recognition. Abrupt changes in tone, repeated euphemisms, unusual urgency, references that only make sense within a certain exchange. Combined with payment timing and ownership links, language becomes one signal among many. The power is not in any single flag. It is in the anomaly chain.

Simon Carver

[thoughtful] “Anomaly chain” is good. Because one anomaly can be innocent. A new company is not a crime. A cousin is not a crime. A rushed renewal is not a crime. But three or four of them aligning... now you’re in story territory, except the machine got there first.

Lachlan Reed

[responds quickly] And that’s where old-school threshold checks fall over. If your rule is just “anything over this dollar amount gets escalated,” mate, you’re playing last decade’s game. Coordinated corruption can stay under every single threshold and still nick the whole trophy. Criminals use the gaps between systems like water finding cracks in concrete.

Jack Burns

[grave] And regulators are evolving. That is the asymmetry many firms do not appreciate. Criminal actors use technology. Regulators increasingly use advanced analytics. Yet many corporations still rely on manual review, static keyword scanning, and periodic due diligence. They are fighting a machine-speed battle with 2005-era controls.

Simon Carver

[questioning tone] Which brings us to the uncomfortable twist. The same system that can spot bribery patterns could also monitor employees, score behavior, scrape internal communications, and decide who looks “risky.” At what point does governance become surveillance with a nicer font?

Lachlan Reed

[uneasy] Yeah. That’s the bit that gives me the wobbles. Today it’s anti-bribery. Tomorrow it’s sentiment analysis on staff messages, predictive HR risk flags, trust scoring -- all dressed up as safety. Same ute, different cargo.

Jack Burns

[firm] Precisely. The same machine that can stop bribery can also become the machine that profiles people. Which is why capability alone is not the measure. Governance matters. Limits matter. Auditability matters. Otherwise the cure adopts the methods of the disease: hidden inference, unchallengeable judgments, diffuse accountability.

Jack Burns

[steady] My position is simple. AI should surface intelligence, not deliver punishment. It should generate leads, correlations, evidence packages, prioritized alerts -- not final guilt. Human beings must remain accountable for decisions, especially where reputation, employment, liberty, or market access are at stake. A hallucinating AI inside compliance is not a software bug. It’s a geopolitical liability.

Simon Carver

[quietly] “Geopolitical liability.” That’s not small language. You mean a false positive isn’t just embarrassing -- it can damage a person, a company, maybe even a cross-border relationship.

Jack Burns

[matter-of-fact] Yes. Freeze the wrong transaction, accuse the wrong intermediary, misread a language pattern, or infer corruption from coincidental ties -- consequences propagate. Banking access can disappear. Careers can end. Regulatory actions can be triggered. The model may be wrong, but the impact is real.

Lachlan Reed

[skeptical] So let me try to say it back. The “10x bank” future -- autonomous monitoring, real-time intervention, continuous compliance intelligence -- sounds unreal on a slide deck. But if nobody can explain why the machine flagged me, then we’ve built a black-box bouncer for the financial system. And good luck arguing with a bouncer made of code.

Jack Burns

[approvingly] Almost. The missing piece is not merely explanation. It is contestability. A human must be able to review, challenge, and override the model. Otherwise dashboards replace judgment, and organizations begin to confuse system output with truth.

Simon Carver

[reflective] That’s the phrase I keep circling: dashboards replace judgment. Because governance theater loves a dashboard. Green, amber, red. A score. A confidence level. It feels objective. It feels clean. But if nobody asks what data was missing, what the model cannot see, who benefits from the scoring logic -- then we’ve automated certainty without earning it.

Lachlan Reed

[softly] And yet... the other side’s no good either. “We didn’t know” used to be a comfy little hiding spot. Bit of plausible deniability, bit of shrugging. In an AI era, regulators may look at a company and say, “You had the tools to know.” That excuse might be cactus.

Jack Burns

[calm] I believe that is the new reality. Ignorance may no longer be defensible where machine intelligence can reasonably expose hidden risk. But the answer is not surrender to machine certainty. The answer is disciplined augmentation: systems powerful enough to detect what humans miss, and governance strong enough to prevent those systems from becoming unaccountable authorities.

Simon Carver

[warmly] That’s where we’ll leave it. If corrupt actors are already using machines to hide the truth, spreadsheets will not save us. But if we hand moral judgment to the machine, we lose something else just as vital. So the question isn’t whether AI enters compliance. It already has. The question is who watches the machine -- and how we defend humanity without automating it away. If this episode gave you something to chew on, like, share, and subscribe.

Lachlan Reed

[friendly] Good one. See you next time, folks.

Jack Burns

[softly] Until then, stay precise.