Don’t Believe Every Bombed Building You See: Inside the Precision of Modern Kill Chains

**Why Precision Migrated to Delivery and Left Selection Behind** _Calm down — take a breath. That bombed-out building on your feed is almost certainly real. What you should not believe is that the image tells you who chose that building, on what evidence, at what confidence, or with what threshold. In an era of sensor fusion, AI-assisted target generation, and curated conflict footage, the thing being hidden is not the destruction. It is the decision._ I am revisiting this piece because its architecture was right and its thesis was wrong, and the correction produces a sharper argument than the one it replaces. The original version argued that much of what circulates as war footage is staged, legacy, or simulated. **A good deal of that survives, and one part of it needs correcting rather than abandoning.** What survives is substantial. Simulated and game-engine footage has demonstrably circulated as authentic combat across multiple conflicts. Legacy imagery is redated constantly. And there is a category of reason for this that critics rarely consider, because it requires knowing something about how operations work: **militaries have excellent motives for not releasing current battlefield imagery.** Real-time footage discloses position, timing, unit, munition, sensor angle, and above all **intent** — publish accurate imagery of where you are working and you have announced where you will be working next. Substituting older or non-specific material is not deception for its own sake; it is elementary operational security, and the alternative is telling the adversary what you know and when you plan to act on it. Anyone treating a released clip as a documentary record of a current operation has misunderstood what the clip is for. So yes: doubt the footage. **Not the existence of rubble somewhere, but the claim that this specific image shows that specific event at that specific time for the reason stated.** Most people confidently narrating a strike could not locate the place on a map, have never stood in the debris, and are reading a caption written by a stranger with an agenda. Common sense applies here exactly as it applies to everything else on the internet, and the person who extends more credulity to an anonymous war clip than to a used-car listing has simply not noticed what he is doing. The one correction worth making is narrow. **The destruction itself is real and at scale** — satellite damage assessment and structural survey establish that independently of any circulating video, and denying it hands away a position that cannot be held and discredits every accurate claim standing beside it. The useful skepticism is not _did anything happen_. It is _what happened, where, to whom, ordered by whom, and on what basis_ — and **none of that is recoverable from an image.** The photograph contains the aftermath and withholds the decision, and the decision is the entire moral content of the event. Which points at the honest thesis, and it is more interesting and considerably more damning than the original. **Precision in modern warfare has migrated almost entirely to the delivery layer while selection remained a classifier problem** — and a precisely delivered munition against a wrongly selected target is not an improvement in ethics. It is an improvement in efficiency. The error does not disappear. It arrives on time. ## The Architecture Is Real Start with what has genuinely changed, because the technical account in the original was substantially sound. The traditional **find-fix-finish** paradigm has been compressed into a cyber-physical loop in which sensor fusion, persistent ISR, and predictive behavioral modeling collapse the interval between detection and engagement. Decisions route through pipelines informed by machine-learned profiles, and the munition is the endpoint of an integrated intelligence process rather than the beginning of a battle. That is a real and enormous change from the era of area bombing and unguided artillery, and its practical effect on **delivery accuracy** is not in dispute. A weapon that lands within a meter of its aimpoint is a categorically different instrument from one with a five-hundred-meter circular error probable. **Distributed lethality** is likewise real, and this is the part public imagination has not caught up to. Swarming systems operating through emergent role reassignment, sharing line-of-sight data and saturating defenses through numbers rather than mass, have moved from research programs to fielded capability, and Ukraine has been the proving ground. But the strategic meaning is missed by anyone still picturing armor columns. **Nobody on the receiving end of modern capability is afraid of tanks.** A tank is visible, slow, expensive, attributable, and killable. What produces genuine fear is the mosquito — a swarm element the size of a bird or smaller, arriving without warning, delivering a lethal micro-payload the way an insect delivers a bite, leaving nothing behind that looks like an attack. The era in which massed formations frightened people is over. The era of being unable to sit outdoors because the sky contains things too small to shoot at has begun, and it is arriving faster than the doctrine to govern it. **Cyber operations as preemptive terrain shaping** is doctrine rather than speculation: infrastructure is mapped and softened before kinetic engagement, and the effects are often camouflaged as malfunction precisely to exploit the latency of disbelief. **Biometric and pattern-of-life identification** — facial recognition, gait, voiceprint, device signature, movement pattern — has made individualized targeting the norm rather than the exception, converting each person into a node in a threat graph rather than a face in a crowd. And the psychological layer is documented on both sides. **AI-assisted influence operations** construct susceptibility profiles from language use and engagement behavior and deliver tailored vectors, which is the same instrumentation as commercial personalization pointed at a different objective. NATO's cognitive-warfare concept names the aim as altering a population's perception of reality in order to affect its decision-making; RAND's analysis of the contemporary propaganda model found a method with **little or no commitment to truth and none to consistency**, whose function is capacity degradation rather than persuasion. Perception management has been formal American policy since NSDD-77 in 1983. ## Where the Original Overreached Three claims need tiering rather than repetition, and marking them is what makes the rest usable. **Neuroweapons, nanodart microdrones, and atmospheric coherence disruption** were presented as fielded capability. They are not. Directed-energy research is real and long-standing; the specific systems described are not documented, and the closest real-world referent — the anomalous health incidents commonly called Havana Syndrome — was assessed by the American intelligence community in 2023 as very unlikely to be the result of a foreign adversary's weapon, with most agencies finding other explanations more probable. _Unsupported._ The category of zero-attribution effects is worth watching. The specific inventory was invented. **Abort-on-detection protocols that scan the impact zone until milliseconds before detonation and self-destruct if a child is present** were presented as standard. That capability, as described, is not documented. What is real sits nearby and is genuinely impressive: **collateral damage estimation methodology**, no-strike lists, pattern-of-life analysis, delayed-fuze and reduced-charge options, and the R9X — a Hellfire variant with deployable blades and no explosive warhead, built to kill a single occupant of a vehicle without destroying it. _Partly real, materially overstated._ **Semiotic harvesting as a deliberate operation** — the claim that traumatic imagery is released following operations specifically to observe who celebrates, mapping ideological volatility for preemption — was asserted as fact. It should be marked as what it is: _plausible, unresolved._ The mechanism requires no deliberate release to function. Real events generate reactions, reactions are harvested by platforms, vendors, researchers, campaigns, foreign services, and security organizations simultaneously, and the resulting behavioral picture exists whether or not anyone designed it. **Effect is not intention**, and the analysis is stronger for declining the unnecessary attribution. ## Malice Is Not the Mechanism Before going further, one model has to be discarded, because it is the largest obstacle to thinking clearly about any of this. The proposition that the United States military, the Department of Defense, and allied forces constitute an apparatus that _wants_ to bomb buildings full of children — that indiscriminate killing is the objective rather than the failure mode — **has no mechanism.** It is not an unkind hypothesis. It is an unworkable one. A strike package does not originate with someone who feels like destroying something. It moves through target development, intelligence corroboration requirements, **positive identification** standards, no-strike list deconfliction, **collateral damage estimation methodology** with numbered severity tiers, proportionality assessment against anticipated military advantage, weaponeering chosen to reduce blast radius, engagement authority that escalates with estimated civilian risk, and **judge advocate review embedded at the strike cell** — lawyers whose professional obligation runs to the law of armed conflict and who have refused strikes. For the malice model to hold, thousands of professionals across multiple services and allied nations would have to be simultaneously corrupt and uniformly silent: those lawyers, the analysts whose job is writing dissenting assessments, the aircrew and operators who have declined missions, and everyone else inside the leak culture that has produced every other disclosure of the last two decades. That is not a theory of institutions. It is a cartoon. The strongest evidence against the malice model is, paradoxically, the **acknowledged failures.** The August 2021 Kabul strike that killed Zemari Ahmadi and members of his family, including children, was investigated, publicly acknowledged as a tragic mistake, and reviewed at the highest levels. Other strikes have drawn internal and congressional scrutiny. The Department of Defense subsequently stood up a **Civilian Harm Mitigation and Response Action Plan** precisely because documented failures demanded an institutional answer. **An apparatus that intends indiscriminate killing does not investigate itself, publish findings, and build a mitigation directorate. It buries.** The existence of the acknowledgment is the disproof of the intent. There is also basic epistemic humility owed to any image of a damaged structure. You do not know what was inside it. You do not know whether it was in military use, whether it held munitions that produced a secondary detonation far larger than the weapon delivered, whether the strike was ours or theirs, whether something ran beneath it, whether it had been evacuated, or whether the footage is even from the conflict it is captioned with. Almost none of that is recoverable from a photograph, and confident moral conclusions drawn from photographs are conclusions drawn from nothing. And the asymmetry that matters most gets flattened in every feed. **For a lawful military, civilian death is a failure. For a terrorist organization, civilian death is the objective.** The café, the bus, the concert, the kibbutz — those are not collateral to the mission, they _are_ the mission, selected precisely because the victims are unarmed. That is the distinction the entire law of armed conflict is built on, and collapsing it — treating a strike that killed civilians as morally identical to an attack that targeted them — is not moral seriousness. It is the erasure of the only thing that has ever separated a military from a massacre, and it is exactly what the adversary's information operations are constructed to produce. It is also worth saying that combatant identification is frequently not the mystery the "algorithm decides who dies" framing implies. A great deal of engagement occurs against people who are actively firing, emplacing devices, or maneuvering under arms — parties who have identified themselves by conduct in the most unambiguous way available. The hard classifier problems are real, and they are a subset rather than the whole. None of which places the system beyond criticism, and here is the reason to insist on all of it. **The malice thesis is the thing that protects the real problem.** So long as the accusation is that the apparatus wants to kill children, the apparatus dismisses its critics without effort, because the accusation is false and everyone inside knows it is false. The criticism that actually lands is narrower, harder, and true: **the selection layer is under-verified relative to its throughput, the thresholds are set by people who do not bear the error, and the recourse is close to nonexistent.** That is a critique a professional can act on. The other is one a professional can ignore, which is why it has changed nothing in twenty years. ## The Other Side Knows Exactly What Game This Is The counterpart to institutional good faith is adversarial sophistication, and the adversary in an asymmetric conflict is not confused about where he can win. This is not a character claim. It is a strategic one, and it follows from arithmetic. A party that cannot prevail kinetically — that has no air force, no armored maneuver capability, no prospect of defeating a modern military in the field — retains one domain where it can win outright: **the perception layer.** Every incentive therefore pushes its investment there. Hamas has operated media production capability as an explicit component of its approach for years, and the broader ecosystem of sympathetic amplifiers, opportunistic accounts, and engagement farms extends that reach enormously without any coordination being necessary. **The weaker party fights where it can win, and it can win in your feed.** The verification record is not ambiguous. Fact-checking desks at BBC Verify, AFP, Reuters, and DW have repeatedly documented footage from Syria, Iraq, Ukraine, and other conflicts recirculated as current; old imagery redated; unrelated industrial accidents captioned as strikes; and — a recurring feature across multiple wars now — **video game footage, most often from Arma 3, presented as authentic combat.** Generative imagery has added a further layer, producing convincing scenes of destruction that never occurred anywhere. The Al-Ahli hospital explosion of October 2023 is the reference case for how fast an unverified attribution can travel: a casualty figure and a cause were asserted within the hour, propagated globally, triggered real-world violence, and were subsequently and substantially revised by multiple independent analyses — after the narrative had already done its work. **A correction that arrives after the riot has not corrected anything.** Two disciplines are required here rather than one. Manufactured and misattributed content is produced by **specific actors and amplifiers**, not by a people, and the overwhelming majority of Palestinians are on the receiving end of this information war rather than operating it — the same distinction that separates Islamism from Islam, and for the same reason: a target set defined by category rather than conduct is a broken target set. And the phenomenon runs in **both directions**; misattributed and manipulated material has circulated from every side of this conflict, and an article claiming otherwise would be doing exactly what it accuses others of doing. But here is why this belongs in an article about kill chains rather than in a separate piece about media. **Adversarial information operations are an attack on the classifier's input layer.** Embedding among civilians, siting military functions in protected structures, staging and recirculating imagery, and disputing every designation after the fact are not separate activities. They are one integrated strategy with two settings: **denial**, which raises the estimated collateral cost until a lawful military declines the strike, and **delegitimization**, which converts any strike that does occur into evidence of atrocity regardless of what actually happened. Both settings exploit the same thing — the fact that a rules-bound force is constrained by rules the adversary does not carry. That asymmetry is not a flaw in the rules. It is what the rules are, and the adversary's entire operational design is built on knowing that you will keep them. ## What "Sophisticated Targeting" Actually Means The public picture of modern targeting is roughly forty years out of date, and the gap between the picture and the reality is the reason so much commentary is unserious. Nobody is dropping area munitions on a neighborhood to reach one person. The relevant capability is **multi-source intelligence fusion producing pattern-of-life models**: persistent overhead observation, signals and electronic intelligence, device and network signature, movement history, associational graph, and behavioral routine, assembled over weeks or months into a model of a specific individual's ordinary day. Where does he sleep, when does he move, who does he meet, which vehicle, which route, which phone, which hours is the building occupied by family and which hours is it not. That model is what makes a strike on a single moving vehicle possible instead of a strike on a block — and it is why capabilities like the blade-only R9X exist at all, a weapon designed to kill one occupant of a car without destroying the car. The honest way to put the scale of it is that **the routine of a tracked individual is frequently known better by the people tracking him than by the people who live with him.** That is not a boast about anyone's virtue. It is a statement about what persistent multi-INT collection does to the concept of privacy in a conflict zone, and it should be unsettling on its own terms. But it settles one question completely: the notion that strikes are guesses, or that the choice of aimpoint is careless, is not a criticism of the system. It is a description of a system that stopped existing decades ago. Which is exactly what makes the real critique the sharp one. **When targeting is this good, error stops being a technical limitation and becomes a threshold decision.** If you can model a man's week, then the residual harm is not what you failed to see. It is what you decided was acceptable, at what confidence, in what hurry — and that decision belongs to people, not to sensors. ## The Real Finding: Precision Without Selection Here is the argument that brings this piece into line with everything else in this body of work, and it is the one the original missed entirely. A kill chain has two distinct accuracy problems and they are almost never distinguished in public discussion. The first is **delivery accuracy** — given a designated aimpoint, how close does the effect land. The second is **selection accuracy** — is the designated aimpoint the correct one, on what evidence, at what confidence, under what threshold. Twenty-five years of investment has produced extraordinary gains on the first and comparatively little on the second, because delivery is a physics problem with a closed-form solution and selection is a **classifier problem operating on adversarial inputs with a base-rate difficulty**. [](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjFkYg6DAYXveiI7_jfhgiATJJgcodXEdanTd61tm3UbvlqS5QqRkY1-Cm7qB6ffUYuD2zT3ahUsvgXaKjZGh-10BpUVOJJPXEDfhipmgtiZHSdo4Qw6HB5B_SSzcSnpLto_gp8Q4LxqpAk2UF35nWDe6yhSmnyVxVifMcwwjfkgVrZO3qaTUP79bQSgrom/s600/modern-warfare-killchains.png) The consequence is uncomfortable in both directions. **The munition became surgical and the target list did not.** A one-meter CEP delivered against a name generated by a model with a permissive confidence threshold, minimal human review time, and an accepted collateral ratio produces a precisely executed error — and at machine tempo it produces many of them, faster than any prior generation of warfare could produce them by accident. The reporting on AI-assisted target generation in Gaza, published by _+972 Magazine_ and Local Call in 2024, described exactly this failure mode: systems generating target lists at volumes far exceeding prior human throughput, with brief human review and pre-authorized collateral parameters. The IDF disputed the characterization of those systems and the specific figures, and the reporting rests substantially on anonymous sources; it belongs at **strongly indicated rather than established**. But the structural claim does not depend on any single account, because it is a property of the architecture: **when target generation is accelerated and target verification is not, the ratio of correct to incorrect designations moves in one direction, and every one of them is delivered accurately.** This is the identical finding this body of work reached about CECOT, where a status classifier running on tattoos and nationality filled a terrorism confinement facility with men who, by the _New York Times_ investigation, largely had no criminal record in the United States or the region. It is the identical finding about Prevent, where thirty-four percent of referrals carry no identified ideology. It is the same arithmetic every time: **the delivery mechanism is never the weak link. The classifier is.** And the moral content of any system of this kind lives entirely in the **cost matrix** — what the score licenses, at what confidence, with what appeal, and against whom the error falls. Which means the operative question about a strike is not the one the footage invites. It is not _how much destruction_. It is **who was designated, on what evidence, reviewed by whom, for how long, against what threshold, with what recourse** — and none of those things are visible in an image of rubble. That is why the title of this piece is still right even though its original thesis was wrong. Don't believe every bombed building you see, not because the building is fake, but because **the image contains no information about the decision, and the decision is the entire moral content of the event.** ## Why the Feed Feels the Way It Does The reason all of this arrives as an emotional emergency rather than as an analytical question has a mechanism, and it is not your weakness. **Real and synthetic footage now occupy the same feed**, and that co-location is what makes verification expensive for everyone. Authentic documentation of an actual strike, an AI-generated composite, an image from a different war, and a video game clip all arrive in the same scroll with the same production values. On the conflict side, that ambiguity is doctrine — the degradation of a population's capacity to sort true from false is a maintained function on every side, not a byproduct. On the entertainment side it is craft, with its own term of art: **TINAG**, _This Is Not A Game_, the deliberate construction of a fiction that denies its own fictionality. A person standing where doctrine-blur and craft-blur overlap has **no reliable local test for which one he is inside.** And the tempo is not accidental either. A system optimizing for information extraction requires informative behavior, behavior is maximally informative under stress, and an environment that must learn about its population will therefore manufacture stakes — which is why intensity escalates monotonically rather than cycling. **The crisis is the probe.** Your reaction to a real atrocity is more diagnostic than your reaction to a fabricated one, because a fabrication mostly filters for credulity while a true event filters for what you want to _do_ about something true. That is the sharper instrument, and it requires nobody to fake anything. ## What Discernment Actually Looks Like The useful response is not generalized suspicion, which produces paralysis and is itself a manipulation outcome. It is narrower and it has been tested. **Technique-based inoculation** works and requires nobody to adjudicate truth. A Cambridge team with Google's Jigsaw built five videos inoculating against specific manipulation techniques — **emotionally manipulative language, incoherence, false dichotomies, scapegoating, and ad hominem attack** — ran them in pre-roll advertising inventory, reached roughly 5.4 million users at about five cents per view, and measured improved discrimination afterward, with the effect holding across the political spectrum. It is source-agnostic. It tells you nothing about what to believe and everything about how you are being moved. Then three habits specific to this domain. **Separate the event from the framing**, because a real strike can be reported in a manipulated frame and a fabricated one in a sober frame, and conflating those is how people end up denying documented events to protect a position. **Ask what is absent from the image**, because the designation, the evidence, the review, the threshold, and the recourse are never in the frame and are always the substance. And **ask what your reaction produces**, because in an instrumented environment your response is data before it is anything else, and the question of who reads it is structural rather than paranoid. ## The Honest Ledger Two things are true simultaneously and any account holding only one of them is choosing a side rather than describing a situation. Modern targeting can be extraordinarily discriminating, and the shift from area destruction to individuated engagement represents a genuine reduction in the _category_ of harm that dominated twentieth-century warfare. Nobody is carpet-bombing a city to reach one commander. That is real moral progress purchased with real engineering. And modern targeting can generate designations faster than any process can verify them, at a tempo where review is measured in minutes, under thresholds set by people who do not bear the error, delivering each mistake with perfect accuracy. **Precision is not accuracy, and accuracy is not justice.** A system that never misses what it aims at, aimed by a classifier nobody audits, is not a humane weapon. It is an efficient one, and the difference between those two words is where every question in this article actually lives. The road ahead does not run through better munitions. It runs through **the same three demands that govern every classifier in this body of work**: designate on conduct rather than status, let trajectory set urgency rather than substitute for evidence, and build the power to clear into the power to flag — because a targeting apparatus that can generate a name and cannot retract one has not solved the problem of war. It has only automated the part that was already easy. * * * [Bryant McGill](https://bryantmcgill.com/about) is a Wall Street Journal and USA Today bestselling author, systems architect, technologist, and strategic advisor, as well as a Congressionally Recognized Ambassador of Goodwill and United Nations–appointed Global Champion. His work spans naval intelligence systems, computational linguistics, artificial intelligence, digital transformation, and civilizational governance architecture. His forward analysis on U.S.–Israel Pax Silica frameworks has appeared in Jewish/Jerusalem News Syndicate (JNS). * * * ## References **Targeting, precision, and the classifier problem** - Yuval Abraham, [Lavender: The AI machine directing Israel's bombing spree in Gaza](https://www.972mag.com/lavender-ai-israeli-army-gaza/), _+972 Magazine_ and Local Call, 2024 — reported and disputed; see IDF response. - UNOSAT, [Gaza Strip Comprehensive Damage Assessments](https://unosat.org/products/). - International Committee of the Red Cross, [International Humanitarian Law and the Conduct of Hostilities](https://www.icrc.org/en/war-and-law/conduct-hostilities). **Cognitive and information layer** - Christopher Paul and Miriam Matthews, [The Russian "Firehose of Falsehood" Propaganda Model](https://www.rand.org/pubs/perspectives/PE198.html), RAND Corporation, 2016. - Jennifer Kavanagh and Michael D. Rich, [Truth Decay](https://www.rand.org/pubs/research_reports/RR2314.html), RAND Corporation, 2018. - NATO Innovation Hub, [Cognitive Warfare](https://www.innovationhub-act.org/). **Anomalous health incidents** - Office of the Director of National Intelligence, [Updated Assessment of Anomalous Health Incidents](https://www.dni.gov/index.php/newsroom/reports-publications), 2023. **Inoculation** - Jon Roozenbeek, Sander van der Linden, Beth Goldberg, Steve Rathje and Stephan Lewandowsky, [Psychological inoculation improves resilience against misinformation on social media](https://www.science.org/doi/10.1126/sciadv.abo6254), _Science Advances_, 2022. ## Collection and ontology routes This article provides the selection-versus-delivery distinction for the [Gamification Collection](https://bryantmcgill.com/collection-gamification). A system can locate, track, and act with extraordinary technical precision while remaining wrong about the identity, intent, or meaning of the selected target. Precision downstream does not repair classification error upstream. Follow [Risk Scoring Systems](https://bryantmcgill.com/wiki/Risk+Scoring+Systems) for the upstream model, [Active Probing](https://bryantmcgill.com/wiki/Active+Probing) for response-based inference, [Playable Systems Literacy](https://bryantmcgill.com/wiki/Playable+Systems+Literacy) for recognizing transmedia and game-like framing, and [Epistemic Injury](https://bryantmcgill.com/wiki/Epistemic+Injury) for the cost of collapsing performance, signal, and fact. [Extremist Elicitation, Remediation, and Classification](https://bryantmcgill.com/article-extremist-classification) develops the classifier; [Never Again and Post-War Gamification](https://bryantmcgill.com/article-never-again-postwar-gamification) places force at the boundary where symbolic substitution has failed. Links to this page [CECOT, Global X, and Offloading Domestic Extremists](https://bryantmcgill.com/article-cecot-global-x) [Gamification](https://bryantmcgill.com/collection-gamification) [Risk Scoring Systems](https://bryantmcgill.com/wiki/Risk+Scoring+Systems) [welcome](https://bryantmcgill.com/articles/welcome)