May twenty twenty-three, southern Lebanon. Hezbollah stages a military drill called Conquest of the Galilee. They've built a full-scale mockup of an Israeli border village — buildings, fences, watchtowers. Militants in tactical gear storm through the replica, breach the mock border fence, simulate taking captives. The whole thing filmed by their own media wing, Al-Manar TV, and released as a propaganda piece. Under open sky. No attempt to hide it.
This is the thing that stops you cold when you see the footage. We live in a world where commercial satellites from companies like Maxar and Planet Labs are capturing imagery at resolutions that can read a license plate. Israel's Unit ninety-nine hundred — that's their visual intelligence unit — is constantly tasking assets over Lebanon and Gaza. The US National Reconnaissance Office runs a whole constellation up there. And yet here's Hezbollah, building a fake Israeli village in a field and running assault drills on it like they're filming an action movie.
Daniel sent us this one. He's pointing at the paradox. Hezbollah, Hamas, ISIS — these groups produce slick training videos on mockups of enemy positions. Hamas before October seventh had a replica kibbutz in Gaza they were drilling on. ISIS ran camps in the Syrian desert. And Daniel's question is basically: in a world this surveilled, do these groups just assume somebody's watching and taking notes from the sky?
The answer is more interesting than the question suggests, because it's not that they assume nobody's watching. It's the opposite. They know they're being watched, and they've built their entire operational model around that certainty.
That's the twist. The thing that looks like a failure of tradecraft is actually the tradecraft.
This isn't a new trick, which is what makes the whole thing stranger. Hezbollah has been running these visible exercises for years. The Conquest of the Galilee drill in May twenty twenty-three was elaborate — MEMRI documented the whole thing. They built a replica of an Israeli military base, complete with watchtowers and perimeter fencing. The footage shows their fighters breaching the mock border, storming structures, dragging away simulated captives. All of it produced and released by their own media arm.
It's not just training. It's a broadcast.
It's a broadcast. And meanwhile, in Gaza, Hamas was doing something structurally similar. They constructed a full-scale replica of an Israeli kibbutz near the border, and over the course of twelve to eighteen months before October seventh, they ran multiple exercises on it. Breaching the fence. Attacking mock civilian structures. Moving through the replica in squad-sized elements.
The surveillance architecture watching all this is not subtle. Unit ninety-nine hundred is specifically tasked with visual intelligence — satellite and aerial imagery analysis over exactly these areas. Maxar and Planet Labs are providing commercial imagery with revisit rates that can hit near-daily coverage over conflict zones. The US National Reconnaissance Office has its own constellation making passes. This is not a sparse information environment. It's saturated.
And that's what makes the surface-level question so obvious that people keep asking it: how do you miss this? You've got satellites. You've got analysts. You've got a militant group building a fake village in a field and running assault drills on it. How is that not a flashing red indicator?
That framing assumes the goal is concealment.
It's not. That's the thing that flips the whole analysis on its head. These groups are not trying to hide the training. They're filming it. They're releasing the footage. They're putting it on television. The question isn't "did intelligence services see it" — the question is "what were these groups doing by making sure they saw it?
The paradox resolves into three possibilities. One: it's a genuine intelligence collection failure — the satellites saw it but nobody connected the dots. Two: it's an analysis failure — the imagery was collected and reviewed, but misinterpreted as routine. Or three: it's strategic — the groups are deliberately showing their hand as part of an information operation, and the visibility is the point.
I think what we're going to find is that it's some of option two and a lot of option three, and that's the part most public discussion gets completely backwards.
Let's walk through what Israeli intelligence actually saw. In the twelve to eighteen months before October seventh, Hamas ran more than twenty visible training exercises on that replica kibbutz. That's not a few. That's a pattern so regular it becomes background noise.
This is where the "they didn't see it" narrative falls apart. They absolutely saw it. Israeli intelligence had imagery of these drills. Analysts reviewed them. The problem wasn't collection — the satellites did their job. The problem was categorization. Each exercise got filed as routine readiness training. Same thing they'd been doing for years. Same formations, same breach drills, same mock structures.
The boy who cried wolf, but the wolf is running the same drill every six weeks and you stop believing it's a real wolf.
That's exactly the dynamic. When you've watched a group run the same exercise twenty times and nothing happened after the first nineteen, your brain stops flagging the twentieth as an imminent threat. It becomes wallpaper. And Hamas understood this. They weren't hiding the training — they were weaponizing its repetitiveness.
The visibility wasn't a mistake. It was the mechanism.
And this connects directly to what Hezbollah was doing with that Conquest of the Galilee drill. MEMRI's documentation of it is striking because Hezbollah didn't just conduct the exercise — they produced it. Multiple camera angles. Al-Manar TV broadcast it as a feature. They wanted Israeli intelligence to see it, and they wanted their own constituency to see that Israeli intelligence was seeing it.
It's a three-audience play. To their recruits and supporters: look at our capabilities, we're ready. To Israel: we're rehearsing for your villages, and you're watching us rehearse, and there's nothing you can do about it. And to the wider Arab world: we are the ones actually preparing for confrontation.
The signaling and the concealment happen simultaneously. What they show — small-unit tactics, breaching techniques, the fact that they've built a replica of your border infrastructure — that's all deliberate display. What they hide — the actual timing, the specific infiltration routes, which units will hit which targets, whether this drill is the real rehearsal or just another decoy in a long string of them — that stays buried.
You're watching a magic trick where the magician shows you the box, shows you the saw, shows you the assistant climbing in, and then dares you to figure out which performance is the one where he actually cuts her in half.
That's a grim but precise analogy. And it works because the cost structure is so lopsided. A satellite constellation costs billions. Analysts, ground stations, tasking priorities, the whole intelligence pipeline — enormously expensive. Building a fake village out of plywood and running some guys through it with cameras? That's cheap. Hezbollah and Hamas have found a way to force sophisticated intelligence services to spend enormous resources staring at what is, in many cases, theater.
Which brings us to ISIS in the Syrian desert, because their approach was different and the difference is instructive. They weren't building mock villages under satellite constellations with near-daily revisit rates. They were in remote desert camps — Al-Tanf, other locations — where coverage was spottier. Lower tasking priority, less consistent passes.
Their threat model was completely different. ISIS wasn't preparing for a conventional cross-border assault on a state military. They were running an asymmetric insurgency — small arms training, IED construction, the kind of thing that doesn't require a full-scale replica of enemy infrastructure. They cared less about state-level satellite surveillance because their operational model didn't depend on massing forces at a border that was being watched constantly.
The visibility strategy maps to the type of fight. Hezbollah and Hamas are preparing for confrontation with a state actor that has overwhelming technical surveillance. They know they can't hide the existence of their training infrastructure, so they don't try. Instead they flood the zone with ambiguous signals. ISIS was fighting a different war against a different set of observers, and their concealment calculus shifted accordingly.
This is what I mean when I say the visibility is not a failure of tradecraft. It's tradecraft adapted to a specific adversary. Hezbollah knows Unit ninety-nine hundred is watching. They know the revisit rates. They know what the commercial providers can resolve. They've built their planning around that knowledge rather than trying to defeat it.
The October seventh intelligence failure, when you strip it down, wasn't "we didn't see the training." It was "we saw the training so many times we stopped believing it meant anything." And the groups knew that would happen.
If they know they're being watched, and they're training anyway, the question shifts from "did we see it?" to "what are they doing with our attention?" And to answer that, you have to understand something these groups have developed that most people don't credit them with: genuine surveillance literacy.
Meaning they've done their homework on what the satellites can actually do.
Not just what they can do — what they can't do. These organizations understand revisit rates. They know that a commercial satellite from Maxar passes over a given location roughly every twelve to twenty-four hours, and that the ephemeris data — the orbital schedule — is publicly available. They know about resolution limits, about spectral bands, about the fact that cloud cover creates gaps. They know when the watcher is blinking.
You can schedule sensitive activity for the gaps. If you know the satellite passes at two in the afternoon, you don't move the thing you want to hide until two thirty.
That's exactly what they do. But it goes further. They mix real preparations with decoy training deliberately. Run the same drill structure twenty times — nineteen of them are theater, one of them is the actual rehearsal for the operation. The analyst watching the imagery sees identical patterns every time and has no way to distinguish which one matters.
Which gets us to the uncomfortable knock-on effect. Post-October seventh, intelligence agencies are doing what bureaucracies always do after a catastrophic miss — they're overcorrecting. Every training exercise now gets flagged as potentially operational. The bar for what counts as a threat indicator has dropped to the floor.
That creates a new signal-to-noise problem that's almost worse than the original one. The US intelligence community has something called "duty to warn" — an obligation to flag indicators of potential violence. When everything looks like an indicator, the warnings become meaningless. Analysts drown in false positives, and the groups know this. They're watching the overcorrection happen and adapting to it in real time.
The asymmetry is brutal when you look at the numbers. A single surveillance satellite costs hundreds of millions of dollars. The ground infrastructure, the analysts, the tasking systems — billions more. And what does it cost Hezbollah to build a fake village and run guys through it with cameras? They've found a way to force the most sophisticated intelligence apparatus on earth to spend enormous resources staring at performance art.
The really uncomfortable implication — the one that keeps analysts up at night — is that the surveillance itself becomes a tool for the observed. The watcher is predictable. The watched knows the watcher's schedule, knows their analysis pipeline, knows what patterns will trigger which responses. At that point, you're not collecting intelligence on them. They're feeding you the intelligence they want you to have.
We saw the same principle in Ukraine. Russian forces building inflatable decoy tanks, running visible training in one sector while the real operation massed somewhere else. Different actor, same playbook. Make the satellite see what you want it to see.
That's the thing that should make us queasy about the whole surveillance-as-solution mindset. We keep adding more sensors, more satellites, more collection — and the groups we're watching just get better at using all of it against us. The panopticon isn't a perfect prison. It's a stage, and they've learned how to perform.
Let's pull out the lenses here, because this is the kind of thing where once you see the pattern, you start noticing it everywhere. First lens: when you watch a propaganda video from one of these groups — the slick production, the storming of the mock village, the captured weapons on display — the instinct is to ask "why are they showing us this?" That's the wrong question.
The right question is "what are they hiding by showing us this?" The visible is always doing two jobs at once. It's projecting a message, and it's drawing your eye away from something else. Every frame of that Hezbollah drill footage is a choice about what to include and what to leave out.
This applies way beyond militant groups. Corporate PR, political campaigns, tech company keynotes — anytime someone is showing you something very deliberately, ask what's being cropped out of the frame. The visible is a distraction from the invisible by design.
Which brings us to the second lens, and this one is for the intelligence professionals listening. Technical collection is necessary. You need the satellites, you need the imagery, you need the revisit rates. But it's not sufficient, and the October seventh failure makes that painfully clear. The hard problem isn't gathering data. It's interpreting intent.
Intent can be deliberately manipulated. That's the thing. You can have perfect imagery of a training exercise and still get the intent completely wrong, because the people running the exercise designed it to be ambiguous.
So the lesson for the agencies is: invest in human intelligence and behavioral analysis, not just more sensors. More satellites won't solve the exercise fatigue problem. More satellites will give you more imagery of exercises you'll misclassify. What you need is people who understand the cultural context, the organizational psychology, the decision-making patterns — the stuff that doesn't show up in pixels.
There's an uncomfortable implication buried in that, which is that the post-October seventh instinct to throw money at technical collection is exactly the wrong response. It's treating the symptom — "we didn't have enough data" — when the disease was "we had too much data and no way to distinguish signal from noise.
That's the third lens, the one for the informed citizen trying to make sense of all this. Be deeply skeptical of "surveillance solves everything" narratives. Ubiquitous surveillance creates its own failure modes. Desensitization to patterns you've seen a hundred times. And the big one — manipulation of the watcher by the watched.
The panopticon is not a perfect prison. It's a stage. And once the people on stage understand the sightlines, the lighting, and where the guards are sitting, they can perform whatever they want you to see.
Here's the final thought that ties it all together. These groups have figured out something counterintuitive about secrecy. The best way to hide a secret is not to conceal it. It's to bury it in an avalanche of other secrets — or to make it look exactly like the thing you see every day.
The most effective camouflage isn't invisibility. It's ordinariness. If your attack rehearsal looks identical to the nineteen routine drills that preceded it, you've hidden it in plain sight. The secret isn't under a rock somewhere. It's on the twentieth page of a report nobody reads anymore because every page looks the same.
That framework — hide the extraordinary inside the ordinary, bury the signal in noise you created yourself — that's portable. You see versions of it in financial fraud, in disinformation campaigns, in corporate obfuscation. Once you know to look for it, it changes how you read almost any deliberate communication.
That's where the framework leaves us — hide the extraordinary inside the ordinary. But here's the question that keeps pushing forward: what happens when the analyst isn't human?
This is where it gets genuinely interesting. Right now, the big push in intelligence is AI-powered analysis tools — computer vision systems that automatically flag changes in satellite imagery. The pitch is that machines won't get exercise fatigue. They won't be desensitized by seeing the same drill twenty times. They'll just...
The obvious counterpoint, which I suspect you're about to make, is that machines have their own failure pattern.
They absolutely do. And the failure pattern are different from human ones, but they're not necessarily better. A computer vision system trained to detect "military training activity" will flag every training exercise, every time. Which sounds great until you realize that's exactly the overcorrection problem we just described — except now it's automated and happening at machine scale.
Instead of twenty reports an analyst might skim, you get twenty thousand alerts nobody can triage.
But the deeper problem — and this is the one I think will define the next decade of this cat-and-mouse game — is that these systems can be gamed in ways that are specific to how they work. Human analysts get bored. Machine learning models get fooled by adversarial examples.
Explain that term — adversarial examples.
It's a concept from machine learning security. You take an image that a model classifies correctly — say, a photo of a training exercise — and you make tiny, carefully calculated changes that are invisible to the human eye but cause the model to misclassify it entirely. The classic example is a panda photo with some noise added that the model suddenly sees as a gibbon with ninety-nine percent confidence.
Applied to satellite imagery, you're saying a group could learn what patterns trigger the AI's "threat" classification and then deliberately create those patterns where nothing is happening.
Imagine Hezbollah arranging some vehicles and equipment in a field in a configuration that they know, from reverse-engineering the system's behavior, will trigger a high-confidence alert. The AI flags it. Tasking priorities shift. Satellites get retasked to stare at an empty field that was arranged to look threatening to a machine. Meanwhile, the real activity happens somewhere else, in a configuration the model was trained to ignore.
That's the adversarial training problem. Not training your own forces — training the enemy's AI to see things that aren't there.
The nightmare scenario is that this degrades trust in the entire intelligence pipeline. If analysts start getting burned by AI false positives that were deliberately induced, they start ignoring the AI. Or worse, they waste enormous resources chasing phantoms while the real indicators get buried in the same avalanche of noise — just now the noise is machine-generated and the groups are controlling the volume knob.
The same fundamental dynamic we've been describing — the watched manipulating the watcher — just gets ported to a new domain. The tool changes, the principle doesn't.
That's the open question I think Daniel's prompt ultimately points toward. We keep building more sophisticated ways to watch, and the watched keep building more sophisticated ways to use our watching against us. AI won't break that cycle. It'll just accelerate it and make the manipulations harder for humans to spot.
Which means the real challenge isn't technical. It's philosophical. How do you design an intelligence system that can't be turned into a tool by the people it's supposed to be watching?
I don't think anyone has a good answer to that yet. And I'm not sure more technology is the path to one.
Which is probably where we should leave it — with the uncomfortable thought that the next generation of surveillance tools might not make us safer. They might just give the people we're watching a bigger stage and a more gullible audience.
Now: Hilbert's daily fun fact.
Hilbert: In the early fifteen hundreds, on New Zealand's South Island, a tardigrade was observed entering a state of cryptobiosis not in response to desiccation, but apparently triggered by the shadow of a passing bird — suggesting a threat-response pathway that bypasses environmental chemistry entirely.
...a tardigrade with an anxiety disorder.
If you've got a weird prompt you want us to explore, send it to show at my weird prompts dot com. We read every one.
This has been My Weird Prompts. I'm Corn.
I'm Herman Poppleberry. Thanks to Hilbert Flumingtop for producing.
We'll be back next week. Try not to think about that tardigrade.
