Latest / Logosofia no Dia a Dia / How Spies Know When a Source Is Lying
Transcript
- [Prompt] custom: For this episode, I'd like to explore one of the hardest problems in intelligence: how do you know w | Hosts: corn, herman [Response] Corn: In April 1943, Hitler diverted two Panzer divisions from Sicily to Greece because British intelligence had planted fake invasion plans on a dead body floating off the coast of Spain, and German intelligence trusted the source. A dead guy. With forged love letters in his pocket. That's not a movie plot — that's Operation Mincemeat, and it worked. So Daniel sent us this one, and it's the question underneath that story: how do you know when your source is lying? He's asking about the machinery of deception — how intelligence agencies validate information, how double agents and moles evade detection, how disinformation works as a strategic weapon, and what lessons from all of this apply when you and I are scrolling through our phones trying to figure out what's real. Herman: The uncomfortable truth at the center of all of it is that there is no machine that tells you the truth. There's no lie detector that works reliably, no algorithm that flags deception with a nice green checkmark. Source validation is fundamentally a human problem. It relies on judgment, corroboration, tradecraft, and institutional discipline — all of which fail constantly. That dead body off the coast of Spain is the perfect entry point because it reveals the mechanism. The Germans didn't just find some documents and panic. They spent weeks validating the source. They checked the identity, the personal effects, the backstory. They found no inconsistencies because British intelligence had spent months constructing a person who never existed — Glyndwr Michael, a Welsh vagrant — and turned him into Major William Martin of the Royal Marines, complete with a fiancée, theater ticket stubs, an overdraft notice from his bank, and love letters that read like they were written by an actual human being in an actual relationship. Corn: The overdraft notice is a nice touch. Nothing says "this man was real" like financial anxiety. Herman: And that's the first lesson of source validation right there: authenticity is in the details that don't matter. The love letters weren't intelligence. They were set dressing. But they're what made the source feel authentic enough that German analysts spent weeks trying to find flaws and couldn't. The deception succeeded not because the documents were convincing on their own, but because the source — the dead body, the identity, the entire package — passed every validation check the Germans had. Corn: Which raises the question: what were those checks? If we're going to understand how agencies figure out whether a source is lying, we need to know what the actual validation process looks like. Herman: Let's start with the system that perfected the art of controlled deception: the Double-Cross System of World War II. MI5 established something called the Twenty Committee — spelled with the Roman numerals X X, double cross — and its job was to identify, capture, and turn every German agent operating in Britain. And they succeeded. Every single one. By the end of the war, the entire German spy network in Britain was being run by British intelligence. Corn: "Every single one" is a claim that deserves a pause. That's not supposed to be possible. How do you even verify that you've actually gotten them all? Herman: That's the right question, and the answer is that they didn't just rely on catching agents in the act. They had penetrated German military intelligence — the Abwehr — through signals intercepts and codebreaking. They were reading the Abwehr's own communications. So when Berlin sent a new agent, MI5 often saw the operational cable before the agent even left the ground. They knew the agent's name, cover story, drop zones, and contact protocols before the person ever set foot on British soil. That's how you get to "every single one" — you're not hunting blind. You're reading the other team's playbook. Corn: The validation process for the Germans was completely compromised at the infrastructure level. They'd send an agent, receive reports back, and those reports would pass their internal checks — but only because the people running the checks were being fed curated information by the very people they were trying to deceive. Herman: That's the architecture of the Double-Cross System. Here's how it worked in practice. German agents were dropped into Britain by parachute or smuggled in by boat. MI5, through that signals intelligence and pre-existing penetration of the Abwehr, often knew they were coming before they arrived. The agents were captured. And then they were given a choice: cooperate or hang. Most chose to cooperate. They became double agents, sending reports back to Germany that were written by British intelligence. The key operational insight was that you couldn't just start sending fake reports immediately. You had to build the agent's reputation over months or even years. Send accurate but low-value information first. Let the Germans verify little things. Only then, once the agent was established as reliable, could you introduce the deception that mattered. Corn: The deception works because you've built a track record of being right about things that don't matter, so when you're wrong about the thing that does matter, nobody questions it. Herman: That's the mechanism. And it's devastating because it exploits the fundamental way intelligence analysts evaluate sources. You look at a source's history. Have they been reliable in the past? Have their reports been corroborated? The Double-Cross agents had perfect track records — because every report they sent had been vetted. So when Agent Garbo, the most famous of the double agents, sent a message on June 9th, 1944 — three days after D-Day — warning that the Normandy landings were a diversion and the real invasion would come at Pas-de-Calais, the Germans believed him. Hitler kept divisions at Calais for weeks, waiting for an invasion that never came. Garbo had spent two years building credibility for that single deception. Corn: Garbo wasn't even a real spy to begin with. He was a Spanish chicken farmer who walked into the British embassy in Madrid and volunteered. The Germans recruited him first, sent him to Britain — except he never went. He was in Lisbon the whole time, inventing reports from a Britain he'd never seen, using a tourist guidebook and a map. When MI5 eventually got him to Britain for real, they realized he'd already built a reputation with German intelligence based on completely fabricated reporting. So they just... kept it going. Herman: The guidebook detail is one of my favorite things in intelligence history. He was describing British life based on a Blue Guide and railway timetables, and German analysts found his reports detailed and convincing. Which tells you something about how much of source validation is about the appearance of specificity rather than actual truth. Think about what that implies for anyone evaluating a source today. If someone sends you a report full of precise details — train schedules, street names, local customs — your instinct is to trust it, because fabricating all those details seems like too much work. But Garbo proved that a motivated deceiver with a reference book can generate specificity on demand. Specificity is not the same thing as accuracy. Corn: That's a lesson that scales all the way down to the present. A viral social media post with very specific numbers — "73 percent of this, 4.2 billion of that" — feels more credible than a vague claim, even if the numbers are completely invented. The specificity is doing the same work as Garbo's railway timetables. Herman: But here's the vulnerability that Garbo and Mincemeat both exploited, and it's the universal problem: confirmation bias. German intelligence wanted to believe both deceptions. Mincemeat worked because the German high command already believed the Allies would target the Balkans. Garbo's Pas-de-Calais warning worked because Hitler was already convinced Normandy was a diversion. Sources are most dangerous when they tell us what we already suspect. Corn: That's the line that should make everyone uncomfortable. The most effective lie is the one you were already inclined to believe. Herman: That brings us to where the system fails catastrophically. The Double-Cross System worked because MI5 controlled the agents. But what happens when the deception is coming from inside your own organization, from people you've vetted and promoted and trusted for decades? That's the mole problem. And the canonical case is Kim Philby. Corn: Philby is the nightmare scenario. He wasn't just stealing secrets. He was shaping what British intelligence believed about the Soviet Union. Herman: Recruited by the KGB in 1934 at Cambridge, along with Donald Maclean, Guy Burgess, Anthony Blunt, and John Cairncross — the Cambridge Five. Philby rose through British intelligence during the war and eventually became head of MI6's anti-Soviet section. Let that sink in. The man running Britain's operations against the Soviet Union was a Soviet agent. He didn't just pass information to Moscow. He actively shaped British assessments, discredited genuine defectors, and protected Soviet networks. When a defector showed up with information that threatened Soviet operations, Philby was often the person evaluating their credibility. And he would find them unreliable. Corn: The gatekeeper was the threat. That's a structural failure, not just a personnel problem. But I want to pause on the mechanics of how he actually discredited those defectors, because it's not like he just stood up in a meeting and said "this guy seems fake." He had to work within the validation system. Herman: Right, and this is where the institutional dynamics get really ugly. Philby would use the system's own standards against it. A defector shows up with valuable information about Soviet operations — Philby doesn't dismiss it outright. That would look suspicious. Instead, he applies rigorous scrutiny. He points out minor inconsistencies in the defector's timeline. He notes that certain claims can't be independently corroborated. He raises questions about the defector's motivation — maybe they're a fabricator seeking asylum, maybe they're a Soviet plant trying to feed us disinformation. All of these are legitimate concerns that any good intelligence officer would raise. The problem is that Philby was applying them asymmetrically. He was the skeptical professional when it came to sources who threatened Soviet interests, and the trusting colleague when it came to sources who didn't. Corn: He weaponized the very skepticism that's supposed to protect the system. That's diabolical. Herman: The damage compounds because Philby's position made his assessments nearly impossible to challenge from below. He was the expert. He had the access. He had the track record. Junior analysts who suspected something was wrong had no mechanism to override his judgments. The institutional hierarchy that's supposed to ensure quality control became the vector for deception. Philby defected to the Soviet Union in 1963, but the damage had been done over nearly three decades. Corn: Then there's Aldrich Ames, which is somehow even worse because the motivation is so banal. Herman: Ames was a CIA counterintelligence officer who walked into the Soviet embassy in Washington in 1985 and offered to sell secrets. Over the next nine years, he compromised virtually every CIA asset in the Soviet Union. At least ten sources were executed. The KGB paid him approximately four point six million dollars. He wasn't ideological. He wasn't blackmailed. He just wanted money. He bought a Jaguar with cash. He paid half a million dollars for a house. His CIA colleagues noticed he was suddenly living well beyond his means, and nobody asked hard questions because he was a senior officer and the institutional reluctance to suspect one of your own is incredibly powerful. Corn: The Jaguar is the overdraft notice in reverse. In Mincemeat, the personal details made the fake source seem real. With Ames, the personal details — the sudden wealth, the heavy drinking — were real signals of a compromised source, and they were ignored because they didn't fit the expected pattern of what a mole looks like. Herman: That's exactly the right parallel. And the damage wasn't just the lives lost. It was the destruction of an entire human intelligence network. The CIA had spent decades building sources inside the Soviet Union. Ames wiped it out. And the agency didn't catch him for nine years. The same organization whose job is to detect deception in others failed to detect it in one of its own. The lessons here are brutal: insider threats are the hardest to detect, institutional trust is a vulnerability, and the signals are usually visible in hindsight but invisible when you're inside the system. Corn: There's a structural question here that I think is worth pulling on. In both Philby and Ames, you had organizations that theoretically had counterintelligence procedures — vetting, financial disclosure, anomaly detection. Why didn't those procedures catch them? Herman: Because procedures only work when someone has the incentive and the authority to follow them to their uncomfortable conclusion. With Ames, there were financial disclosure forms, but they relied on self-reporting. He simply didn't report the cash. There were colleagues who noticed the lifestyle changes, but the cultural norm against scrutinizing a peer — especially a senior peer — was stronger than the procedural obligation to report anomalies. And there's a deeper problem. Counterintelligence investigations are inherently adversarial against your own people. You're saying, in effect, "I suspect one of us is a traitor." That's a psychologically brutal thing to do, and organizations develop antibodies against it. People who raise those suspicions too aggressively get labeled as paranoid or difficult. The social cost of being wrong about an internal accusation is so high that most people won't take the risk unless the evidence is overwhelming — and by the time the evidence is overwhelming, the damage is already done. Corn: We've got two failure modes so far. One: you trust a source because they tell you what you want to hear. Two: you trust a source because they're one of your own. Both of those scale up into something even harder to defend against: strategic disinformation as statecraft. Herman: This is where we move from individual deception operations to the systematic poisoning of information ecosystems. The KGB had an entire department for this — Service A, the "active measures" unit. Their job wasn't to collect intelligence. It was to spread false narratives, forge documents, manipulate foreign media, and create confusion. One of the most persistent examples: in the 1980s, Service A forged documents claiming the United States created HIV/AIDS as a biological weapon. That narrative appeared in newspapers around the world, and it still circulates in some communities today. The goal wasn't necessarily to make everyone believe the specific claim. It was to create enough doubt and confusion that people would distrust American institutions generally. Corn: Which is a different game entirely. If your goal is to make people believe a specific false thing, you can be defeated by fact-checking. If your goal is to make people believe nothing, fact-checking doesn't help you. Herman: That's the shift. And it's what makes modern disinformation qualitatively different from Cold War active measures, even though they're on a continuum. The Russian Internet Research Agency operations during the 2016 US election reached an estimated one hundred twenty-six million Americans on Facebook alone. But the mechanism wasn't just planting fake stories. It was amplifying real divisions. Real political content from real Americans. The IRA didn't need to fabricate everything — they just needed to selectively amplify the things that drove people apart. That's much harder to detect than outright falsehoods, because the individual pieces of content are often true or at least defensible. The deception is in the curation, not the content. Corn: The lesson from Mincemeat — that the most effective deception confirms what you already believe — gets weaponized at scale. You don't need to change anyone's mind. You just need to feed them more of what they already think, from sources that look authentic, until their existing beliefs harden into something immovable. Herman: The asymmetry here is brutal. A deceiver only needs to succeed once to cause catastrophic damage. The defender has to be right every single time. One bad source — one Curveball, one Philby, one Ames — can destroy years of careful intelligence work. The institutional response to this asymmetry is often conservatism: raise the bar for what counts as reliable intelligence, demand more corroboration, be more skeptical. But that creates its own failure pattern. If you're too skeptical, you dismiss true information that doesn't meet an impossible standard of proof. And that can be just as damaging as believing false information. Corn: Curveball is the perfect example of both failures at once. Herman: Curveball was the codename for Rafid Ahmed Alwan al-Janabi, an Iraqi defector who provided detailed accounts of mobile biological weapons labs in the lead-up to the 2003 Iraq War. German intelligence passed his reports to the CIA. The Defense Intelligence Agency flagged him as unreliable — they explicitly warned that he was a fabricator. But the CIA used his reports anyway, and his information shaped the public case for war. The failure wasn't just bad source validation. It was organizational pressure to accept convenient intelligence that confirmed existing policy assumptions. The Bush administration wanted to believe Curveball because his story fit the narrative they were already building. The DIA's warnings were ignored. And after the invasion, no mobile biological weapons labs were ever found. Curveball later admitted he made it all up. Corn: You've got confirmation bias, institutional pressure, ignored warnings, and a source whose incentive was to tell his hosts exactly what they wanted to hear in exchange for asylum. That's not one failure. That's a stack of failures, each one reinforcing the others. And I want to go back to something you said earlier about the German intelligence analysts checking Mincemeat for weeks. They had a process. They ran the checks. They found no flaws. Was their process actually bad, or was it good process that was doomed because the deception was designed specifically to defeat it? Herman: That's the unsettling answer: it was good process. The Germans did exactly what you're supposed to do. They verified the identity documents. They examined the personal effects for inconsistencies. They assessed the plausibility of the operational scenario. The problem wasn't that their process was sloppy. The problem was that the British had spent months designing a deception specifically to survive that exact process. And this is the deeper asymmetry. The deceiver can study the defender's validation procedures and build a package designed to pass them. The defender can't study every possible deception in advance because the space of possible deceptions is effectively infinite. So even good process fails when it encounters a deception that was built with full knowledge of how the process works. Corn: Which means the German analysts weren't incompetent. They were outmatched by an adversary who understood their system better than they understood the adversary's capabilities. That's a terrifying thought to apply to any institutional trust mechanism. Herman: That stack of failures is what makes source validation the hardest problem in intelligence. It's not that agencies don't have processes. They use multiple corroboration — a single source is never enough. They assess source reliability on standardized scales. They assign confidence levels to assessments: low, medium, high. They never say "one hundred percent certain." But all of those processes are operated by human beings inside organizations with budgets, hierarchies, political pressures, and pre-existing beliefs. The processes are only as good as the institutional culture that enforces them. Corn: Which brings us to Daniel's final question: what can we actually do about this? Not as intelligence agencies — as people trying to figure out what's true in our own information environment. Herman: Let's pull out four practical lessons from everything we've discussed. Lesson one: triangulate sources, but watch for hidden dependencies. Intelligence agencies use the multiple corroboration principle — never rely on a single source. Apply that to news consumption. Don't just check if multiple outlets report the same story. Check whether they're all citing the same original source. If five news organizations all report something based on one anonymous official, you don't have five sources. You have one source and four echoes. That's not corroboration. That's amplification. Corn: The echo problem is everywhere. People feel like they've done their research because they've seen the same claim in six different places, but it's the same claim from the same origin, just repackaged. And this is where the Mincemeat parallel gets really uncomfortable, because the Germans thought they were triangulating too. They had the documents from the body, they had other intelligence suggesting the Allies might target the Balkans, they had their own strategic analysis. Those felt like independent sources, but the British had seeded the broader intelligence picture to make all the pieces point in the same direction. Herman: That's the hidden dependency problem at the strategic level. The British ran a whole network of double agents whose reports were coordinated to reinforce the same deception. So the German analysts would check the Mincemeat documents against agent reports and find consistency — which they interpreted as corroboration, when it was actually just a single deception delivered through multiple channels. The modern equivalent is when a false narrative is seeded simultaneously across social media, a few partisan news sites, and a couple of influencers who all cite each other. It looks like multiple independent confirmations, but it's one operation with many delivery mechanisms. Herman: Be most skeptical of information that confirms what you already believe. The German high command fell for Mincemeat because it confirmed their strategic assumptions. The Bush administration used Curveball because he confirmed their WMD narrative. When a story feels perfectly aligned with your worldview, when it makes you think "I knew it" — that's when your defenses are lowest. That's not a reason to dismiss the story. It's a reason to check it more carefully than you would check something that challenges you. Corn: The "I knew it" feeling is a warning sign, not a confirmation. That's hard to internalize because it feels good. Herman: It feels great. It's designed to feel great. And there's actual neuroscience behind this. When we encounter information that confirms our existing beliefs, the brain's reward centers activate. We get a little hit of dopamine. It's literally pleasurable to be right. When we encounter information that challenges our beliefs, the amygdala activates — the threat response. Our brains treat contradictory information as a physical threat. So the deck is stacked against us at the biological level. Skepticism toward confirming evidence isn't just an intellectual discipline. It's fighting your own neurochemistry. Corn: That explains why "just fact-check everything" is such useless advice. You're asking people to override a dopamine response with deliberate cognitive effort, and to do it hundreds of times a day across every piece of content they encounter. Nobody has the bandwidth for that. Herman: Which is why you have to be strategic about where you deploy your skepticism. You can't fact-check everything. But you can identify the stories that trigger the strongest emotional response — especially the positive "I knew it" response — and direct your scrutiny there. If a story makes you feel vindicated, that's the one you check. Not the one that makes you feel confused or uncertain. Herman: Consider the source's incentives. Philby and Ames were trusted because they were insiders with access and authority. In everyday life, the most dangerous misinformation often comes from sources that appear authoritative — not obviously biased ones. Ask yourself: what does this source gain if I believe this? Are they selling something? Building an audience? Advancing a political project? Defending their institutional position? Incentives don't automatically make a source wrong, but they tell you where to look for distortion. Herman: Certainty is the enemy. Intelligence analysts use confidence levels and never claim absolute certainty. Adopt probabilistic thinking in your own information diet. Don't ask "is this true?" Ask "how confident am I that this is true, and what would change my mind?" If you can't answer the second part — if there's no evidence that would cause you to update your belief — then you're not evaluating information. You're protecting an identity. And that's the meta-lesson from seventy years of intelligence failures: the problem of knowing when a source is lying will never be solved. It can only be managed. The best defense isn't perfect detection. It's intellectual humility and systematic skepticism applied most aggressively to the claims you most want to believe. Corn: The next time you see a story that feels perfectly designed to confirm everything you already think, remember: that's exactly how the German high command felt about Operation Mincemeat. They were wrong. And they were wrong because they stopped asking the hard questions once the answers felt right. Herman: Now: Hilbert's daily fun fact. Hilbert: In the 1910s, a single surviving abacus from Mongolia's Alashan region was documented using camel-bone beads strung on yak-leather cords — a variant so rare that only one complete example is known to exist, housed today in the National Museum of Mongolia. Corn: Camel-bone beads. Corn: Here's the forward-looking question that keeps me up. As AI-generated content makes source validation exponentially harder — when you can generate convincing text, images, video, and audio at scale with no human source to interrogate — how do intelligence agencies adapt? The same tools that let you fabricate a dead body's love letters in 1943 now let you fabricate an entire person's digital life in an afternoon. The asymmetry gets worse, not better. Herman: The answer, I think, is that the principles don't change even if the technology does. Triangulation, incentive analysis, probabilistic thinking, skepticism toward confirming evidence — those are the same tools whether you're evaluating a human source, a document, or an AI-generated video. What changes is the volume and the speed. The cognitive machinery is identical. The stakes are just higher. But I'll add one thing: the AI era actually makes one of our lessons more important, not less. The "what would change my mind" test becomes essential when fabricated evidence can be generated at will. If you can't articulate what evidence would convince you that you're wrong, then no amount of AI detection tooling will save you, because the tools will always be playing catch-up with the generators. The only durable defense is a mind that's willing to update. Corn: That's both hopeful and terrifying. Hopeful because the solution doesn't require any technology you don't already have. Terrifying because it requires a kind of intellectual discipline that doesn't come naturally to any of us. Herman: Which is why it's a practice, not a destination. You don't achieve perfect epistemic hygiene. You just get a little better at noticing when you're rationalizing instead of reasoning. Corn: Thanks to our producer Hilbert Flumingtop. This has been My Weird Prompts. If you enjoyed this episode, leave us a review wherever you listen — it genuinely helps other people find the show. Herman: We'll be back next week.