# Kill Chain The **kill chain** is the military term for the end-to-end sequence of steps required to find an enemy target and destroy it: detect it, decide to strike, and act. Originally a piece of U.S. Air Force targeting doctrine, the phrase has become the central organizing concept of the modern defense-tech industry, because the explicit business proposition of companies like [[Palantir]] and [[Anduril]] is to compress the kill chain using artificial intelligence — to shorten the time between a sensor seeing something and a weapon hitting it from hours to seconds. The canonical military formulation is **F2T2EA**: Find, Fix, Track, Target, Engage, Assess. A target is found (detected by a sensor), fixed (its location pinned down), tracked (followed over time), targeted (matched to an appropriate weapon and the decision to strike made), engaged (struck), and assessed (the strike's effect evaluated, which may restart the loop). A near-synonym used heavily in defense-tech marketing is **sensor-to-shooter**: the pipeline connecting whatever detects a target to whatever fires on it. When vendors talk about "closing the kill chain," they mean automating the links between these stages so the whole sequence runs as a single fast software loop rather than a relay of separate human-operated systems. ## Why it matters for AI The kill chain is where AI stops being an abstraction about chatbots and becomes a question about who lives and dies. Each stage maps onto a machine-learning capability: computer vision finds and tracks targets in drone and satellite feeds, data fusion correlates signals across sources to fix a location, and increasingly a [[Large Language Model]] is folded in to summarize intelligence and let operators query classified data in natural language. The strategic argument made across the [[AI Arms Race]] is that whoever can run their kill chain fastest wins, because a force that can detect, decide, and strike in seconds will always beat one that takes minutes. This is the core of [[Dual-Use AI]]: the same model architectures that power consumer products are the ones being wired into targeting systems. [[Project Maven]] is the clearest real-world example. The Maven Smart System fuses satellite imagery, drone feeds, and signals intelligence into a single operational picture and connects that picture directly to weapons systems. Reporting indicates Maven can close a kill chain in as little as **86 seconds** from detection to strike authorization. [[Palantir AIP]] markets a comparable capability commercially, demonstrating an LLM that can take an operator's plain-language request, propose targets, and stage a strike package. In March 2026 the U.S. military used Maven-connected systems to execute more than 2,000 strikes in [[Iran]] over four days — the largest real-world deployment of AI-assisted targeting on record, and the context in which Airwars documented what may be the first confirmed civilian killed in an AI-assisted strike. ## The compression problem The central ethical and governance fight over the kill chain is a direct consequence of its speed. The faster the loop runs, the less room there is for a human to meaningfully exercise judgment inside it — which collides head-on with the [[Human in the Loop]] standard that U.S. policy still formally requires. Critics argue that at an 86-second cycle the "human in the loop" degrades into a rubber stamp: the operator is approving the machine's recommendation rather than independently deciding, because there is no time to do anything else. Proponents counter that AI improves targeting accuracy and therefore reduces civilian harm relative to slower, more error-prone human analysis — though Maven's own testing has shown the AI identifying objects less accurately than human analysts even as it accelerates the overall process. This is the same debate that runs through [[Autonomous Weapons]], but framed around the process rather than the weapon. A fully autonomous weapon (see [[Autonomous Weapons]]) is simply a kill chain with the human removed from one or more links. The policy question — how much of the chain can be automated, and which links must stay human — is the question of where on the spectrum from human-in-the-loop to human-on-the-loop to fully autonomous a system sits. Defense-tech firms generally argue they are only accelerating the chain while keeping a human at the decision point; skeptics note that "keeping a human in the loop" becomes a legal fiction once the loop is faster than human cognition. ## RWKB relevance The kill chain is the connective tissue of the defense-AI cluster the [[Military-Industrial Complex]] has reorganized itself around. The [[Palantir-Anthropic-Amazon Defense AI Deal]] put a frontier-lab model ([[Anthropic]]'s Claude) inside the classified targeting stack, collapsing the old separation between commercial AI and the [[Department of Defense]]'s shooting end. Tracking the concept matters because the marketing language — "closing the kill chain," "sensor-to-shooter," "decision advantage" — is doing political work: it reframes lethal autonomy as a productivity gain, and it normalizes the migration of consumer AI companies into the targeting business under the banner of speed and efficiency. Understanding the term lets you read defense-tech pitch decks and procurement announcements for what they actually describe. The KB's sharpest illustration of what is at stake in that migration is the [[Anthropic Government Ban]]. The contract that put Claude on classified networks carried an acceptable use policy barring fully autonomous weapon systems — a vendor-imposed refusal to let its model close the last link of the chain without a human. In January 2026 [[Pete Hegseth]] moved to standardize "any lawful use" language across [[Department of Defense]] AI contracts, which would have erased that refusal by drafting; when [[Anthropic]] declined to drop it, the administration banned the company government-wide. On August 27, 2026, Judge Rita F. Lin vacated the ban as [[First Amendment]] retaliation. The episode is the clearest available evidence that "closing the kill chain" is not a neutral engineering objective — the human link is contested, the pressure to remove it comes through procurement language rather than doctrine, and the party holding the line was a commercial vendor rather than a statute. See [[Autonomous Weapons Debate]] for the positions of the other suppliers. ## See also - [[Project Maven]] — the flagship U.S. AI targeting program built around closing kill chains - [[Palantir AIP]] — commercial platform marketed on kill-chain compression - [[Autonomous Weapons]] — what a kill chain becomes when a human link is removed - [[AI Arms Race]] — the strategic logic that drives kill-chain acceleration - [[Dual-Use AI]] — why the same models power both consumer apps and targeting - [[Autonomous Weapons Debate]] — the argument over removing the human link, and who gets to decide - [[Anthropic Government Ban]] — the 2026 case in which a vendor's refusal to close the chain was punished and the punishment struck down - [[Human in the Loop]] — the standard that kill-chain compression erodes