We kill people based on metadata.
In 2014, former NSA and CIA director Michael Hayden uttered these words with a straight face. He was not being apologetic but rather bragging about his agency's capabilities. His comment admits a difficult truth about why governments upgrade their surveillance apparatus. They seek to spy on their population, but the ability to kill “enemies of the state” is the ultimate expression of all surveillance.
Most people do not understand how metadata—“data about data”—is used in government “kill chains,” nor do most realize how much that capability is growing. Though such technology is putatively for foreign threats, these kill chains can eventually be turned against any domestic population.
With this threat escalating, we need solutions that invalidate it, which the crypto and cypherpunk communities are already deploying. But before looking at them, we must understand how the US government and military exploit metadata analysis not only to spy but also to develop these kill chains for targeted assassination. Then, we can determine what countermeasures we can muster to defend ourselves and secure a better future with a new kind of “oblivious” institution.
What Is a Kill Chain? The F3EAD Targeting Cycle Explained
First, what exactly is a kill chain?
It is the series of steps a government agency takes in the process of surveilling a person or persons and then green-lighting their assassination via drone bombing. The targeting framework used by US military and special operations units to hit targets is called F3EAD: Find, Fix, Finish, Exploit, Analyze, and Disseminate. The Finish phase is the visible attack that people can see and scrutinize. The Find and Fix phases focus on identifying and locating the target. They rely on preexisting surveillance technologies deployed by the CIA and the NSA, much of the same architecture that has also been used on US citizens.
How the NSA Turns Metadata Into a Pattern of Life
Snowden revealed this program in 2013. He disclosed the existence of XKeyscore, an NSA system that lets analysts search collections drawn from a variety of sources: emails, chats, browsing histories, and all associated metadata. The who, when, where, and duration are all indexed and retrievable, even if the content remains private. Governments no longer require content for ongoing surveillance. What they prize is metadata. The message's content is irrelevant by comparison.
What they get from accumulating this metadata is unparalleled pattern recognition. They use these newfound patterns to determine their targets’ relationships, daily habits, and documented connections. It provides a level of granularity that content alone cannot achieve.
Location data compounds this knowledge and makes the graphing process relatively straightforward. Cellphone towers generate records. Military intelligence agencies use Stingray-class devices to geolocate phones. Together, these technologies reveal the “pattern of life.” This is the surveillance piece of the kill chain logic. It allows intelligence agencies to track their targets for months, sometimes years, before pressing the red button.
The Red Button: How Drone Strike Decisions Are Made
The US military and intelligence community maintain two types of strike decisions: the personality strike and the signature strike.
The personality strike is precise. It relies on more than metadata. The individual is confirmed through multiple sources, including live human intelligence. A team assembles a dossier, pinpoints the individual, and authorizes a strike.
How the Soleimani Drone Strike Was Targeted
We have a documented example of how this “kill chain” operates from find to finish.
In 2020, a US-led drone strike killed Qasem Soleimani, commander of the IRGC’s Quds Force. The operation combined human intelligence with pattern-of-life analysis. That analysis was built over years. US and allied agencies had mapped his movements through signals intelligence for well over a decade. In the end, JSOC liaisons worked with Israeli counterparts to track his cellphone patterns. The Israelis held his numbers and handed them to US agencies, who traced his active handset to Baghdad even though he had switched phones three times in the six hours before boarding.
Long story short, the CIA knew Soleimani would be there. They wanted to confirm his whereabouts through human assets on the ground. Informants at the airport in Damascus tipped off the CIA when his plane took off, and American assets at Baghdad's airport confirmed its arrival. After he left via his convoy, an MQ-9 Reaper drone circling in the sky above launched Hellfire missiles, killing him, Abu Mahdi al-Muhandis, deputy commander of Iraq's Popular Mobilization Forces, and five others.
Signature Strikes: Probabilistic Killing by Metadata
The other type of strike is more alarming. It is called a signature strike, and it relies solely on metadata and pattern-of-life behavior. The operators select individuals to target based on their behaviors and activities. It is a speculation-based hit; in reality, it is just targeting phones and information. The analysts do not know if it’s the actual target. There is no eyewitness to verify that the signature matches a specific person. When a pattern crosses a threshold, the team’s ranking officer authorizes the hit. The aggregate pattern must match a predefined lethal signature. This is called probabilistic killing. It is derived from unsubstantiated surveillance patterns.
All of this information was once top secret but was eventually leaked.
The practice was confirmed in the “Drone Papers,” leaked by Daniel Hale and published by The Intercept in 2015. The papers show that during five months of Operation Haymaker in northeastern Afghanistan, roughly 90% of the people killed were not the intended targets. Across the broader campaign, more than 200 people were killed. Only 35 of them were the intended targets. These individuals are classified as EKIA, “enemy killed in action,” even though they were never confirmed as combatants. It is a way for the US military to make its errors appear palatable to the media. More on this shortly.
Palantir and Surveillance as a Service
These capabilities have been expanded in recent years by tech companies like Anduril Industries, but especially by Palantir. Think of Palantir as the epitome of surveillance weapon technology. Its Gotham software is embedded in tech operated by the CIA, the Department of Defense, ICE, and various domestic police departments. Palantir was co-founded and funded in 2003 by Peter Thiel. The CIA, via In-Q-Tel, also invested early.
Palantir’s goal as a software company is to expertly combine intelligence inputs. It integrates satellite imagery, financial transaction data, biometric data, and social network mapping, aggregates it all, and produces a detailed social graph: an offering we can call surveillance-as-a-service. This kind of analysis enables Palantir to piece together a person’s relationship network based on interaction activity. Any seemingly innocuous piece of info is metadata that can be gobbled up for social graphing. Palantir is now thoroughly embedded in the targeting and data-collection regimes of various government agencies.
Project Maven: The Pentagon's AI Targeting Platform
The Pentagon released its targeting program, Project Maven, in 2017. They intended it to be their computer vision and drone-footage processing program. It was eventually transformed into an AI-targeting platform. At the outset of that program, the Pentagon hired Google to oversee it, but they left in 2018. Google employees took issue with the idea of supporting an autonomous weapons platform. Approximately 4,000 employees signed a petition to reject it. Seeing the opportunity, Palantir signed a contract to take over.
Alex Karp and the Case for Software-Enabled Hard Power
In 2025, Palantir CEO Alex Karp and Nicholas Zamiska published The Technological Republic: Hard Power, Soft Belief, and the Future of the West. The book is a “sweeping indictment of the West’s culture of complacency, arguing that timid leadership, intellectual fragility, and an unambitious view of technology’s potential in Silicon Valley have made the U.S. vulnerable in an era of mounting global threats.” Instead of working to solve national problems, Karp and Zamiska complain, Silicon Valley has chosen to focus on consumer-facing applications.
Their argument, in part, is that tech entrepreneurs and developers are obligated to the state: “Silicon Valley owes a moral debt to the country that made its rise possible.” Its engineering professionals and elite carry “an affirmative obligation to participate in the defense of the nation,” saying that soft power is finished, that free and democratic societies need to prevail with hard power, and that “hard power in this century will be built on software.” By advancing this thesis, of course, they are also defending their own company and its objectives.
Is software-enhanced weaponry a critical piece of future hard power? Is it a question of whether the tools will be built or merely who will build them?
Some argue the US’s foes will not debate the morality of producing surveillance or weapons platforms. They will just do it. Thus, the argument for such programs is a zero-sum arms race: if the US does not build automated targeting and precision weaponry, a bad person or persons will.
Is this a self-defeating argument, however? Karp appears ready to gaze into the abyss so long that it gazes right back, which is exactly what has happened with him and the US surveillance industrial complex. Is there any such thing as a “good guy” who will do the right things with weapons, AI, or weapons platforms? Effectively, whenever an individual or group has catastrophe weapons or large caches of arms at their disposal, there is always a concern that they will use them for ill or harm. And currently, judging by the state of the current system, its AI arsenal and machinery seem ready-made for misuse.
The Bureaucracy of Assassination: Civilian Deaths as Statistics
The most troubling problem with all autonomous warfare tech is that by relying on probabilistic killing and aggregated pattern analysis, one must accept potentially random outcomes. The military and government can only hope they hit the intended target.
However, the truth may be even more ominous. The Drone Papers tell a story of purposeful design behind this inadvertent killing. The way the system is constructed, whoever gets killed is of no consequence as long as a designated target is assassinated occasionally.
For example, the US military and its allies have hit many civilian locations, seemingly without remorse or consequence. These include: the MSF hospital in Kunduz, Afghanistan, obliterated by an AC-130 gunship in 2015; the Dahyan school bus in Yemen, hit in 2018 by a Saudi-led coalition airstrike using a US-supplied bomb; the Al-Shifa pharmaceutical factory in Sudan, leveled by US cruise missiles in 1998; and the Shajareh Tayyebeh elementary school in Minab, southern Iran, hit on February 28, 2026, in a strike that killed at least 165 children, teachers, and parents.
The government might have investigated itself for these strikes, but it typically finds no wrongdoing. The government gives itself allowances for noncombatant casualties, and they usually simply count accidental deaths as a bureaucratic statistic. In this vein, a system that consistently produces a specific outcome appears specifically designed to do so. In other words, the system appears designed to kill innocent bystanders, or at least has a tolerance for a certain amount calculated into it.
Signature Strike Logic Comes Home: Predictive Policing and ICE
The signature strike logic does not only apply to assassinations in foreign lands; it can be deployed domestically against its own residents and citizens. The signature strike logic makes it far easier to dehumanize targets as mere patterns and bits of data. Local police and ICE agents are already using this technology's surveillance capabilities for predictive policing.
In many ways, what is happening reads like Philip K. Dick's Minority Report. There is a framework in place that predicts who will commit a crime and judges those people without due process, much less a trial by their peers. Given the frequency with which fiction becomes reality, is it so hard to imagine that individuals may soon have their own “minority report,” for which the government claims its only viable course of action is targeted assassination?
The very existence of this technology is a serious cause for concern. It has already been abused, and it will undoubtedly continue to be. Our purpose here is not fearmongering, however. Rather, by understanding what is happening, we can better address it.
Breaking the Kill Chain: Why Metadata Minimization Works
The kill chain architecture has a major dependency: metadata aggregation. If large numbers of people remove their metadata signals, pattern-of-life analysis becomes harder to generate. The surveillance machinery, specifically the “find” phase of F3EAD, is thus rendered significantly more difficult.
Whether that would cause the government to rethink its controversial surveillance programs is an open question. Either way, there are techniques and technologies that anyone can use to enhance their privacy or reduce their metadata footprint. And those will be the subject of Part 2.
