AI Weapons

Lavender: The AI That Chose Who Dies in Gaza

In April 2024, +972 Magazine and Local Call published a devastating investigation based on interviews with six Israeli intelligence officers who worked with the Lavender system during the Gaza war. Their revelations are among the most disturbing documents in the history of automated warfare.

How Lavender works: The system analyzes surveillance data — phone records, social media activity, location data, communication patterns, group affiliations — and assigns each person in Gaza a numerical score from 1 to 100 indicating the probability they are a militant. Anyone above a certain threshold is automatically added to the kill list.

The human rubber stamp: Sources described the “human review” process as a formality. An officer would verify that the target was male and that the system hadn't flagged any obvious errors. This process took approximately 20 seconds per target. One source said: “I would invest 20 seconds for each target at this stage, and do dozens of them every day.”

The civilian cost: Intelligence sources revealed the IDF accepted that for every low-ranking Hamas operative killed, up to 15–20 civilians could be killed as “collateral damage.” For senior commanders, the ratio was even higher. One source said the policy was to bomb suspected militants in their homes — at night, when their families were present — because it was easier to locate them there.

The error rate: Sources estimated Lavender had an error rate of approximately10% — meaning roughly 1 in 10 people flagged as militants were misidentified civilians. With 37,000 targets, that means approximately 3,700 people were marked for death by mistake. The system's errors were considered acceptable.

Drone Swarms: The Future of War Is Cheap and Disposable

The Pentagon's Replicator initiative represents a paradigm shift in military thinking. Instead of expensive, exquisite platforms like $100M F-35s and $13B aircraft carriers, the future of war may be thousands of cheap, AI-enabled, disposable autonomous systems — drone swarms that overwhelm defenses through sheer numbers.

The concept is simple: a $2 billion aircraft carrier can be sunk by a swarm of 1,000 drones costing $10,000 each ($10 million total). An air defense system designed to track and engage dozens of targets is overwhelmed by hundreds of autonomous drones attacking simultaneously from different directions. The economics of defense collapse.

China is leading in drone swarm technology. In 2022, the PLA demonstrated a swarm of 200+ drones operating autonomously — sharing targeting data, coordinating attacks, and adapting to losses without human intervention. The US Replicator initiative is explicitly designed to counter this capability with American swarms.

The result is an autonomous weapons arms race with no rules, no treaties, and no brakes. Both sides are developing AI-controlled swarms of lethal autonomous weapons, each driving the other to remove human oversight in the name of speed. In a swarm-vs-swarm engagement, the side with a “human in the loop” loses because human decision-making is too slow. The logic of competition drives both sides toward full autonomy — machines fighting machines, at machine speed, with human life as collateral.

The Libertarian Case: No Algorithm Has Due Process

The Fifth Amendment to the United States Constitution states: “No person shall be... deprived of life, liberty, or property, without due process of law.” An AI system that generates kill lists based on phone records and behavioral patterns is the antithesis of due process. There is no trial, no defense, no jury, no appeal. An algorithm assigns a number, a human rubber-stamps it in 20 seconds, and a bomb falls.

The government claims these protections don't apply to foreign nationals in war zones. But the US killed American citizens Anwar al-Awlaki and his 16-year-old son by drone strike with no trial. The legal reasoning — that “due process” doesn't require “judicial process” — is an Orwellian destruction of the concept. If the executive branch can unilaterally decide that due process has been satisfied, then the right is meaningless.

AI weapons accelerate this erosion. When a human decided to kill someone, there was at least the possibility of moral reflection, doubt, mercy. When a machine generates a target list of 37,000 people and a human spends 20 seconds “reviewing” each one, the decision to kill has been outsourced to an algorithm. The human is decoration.

The libertarian principle is clear: no government should have the power to kill people based on algorithmic predictions. Not foreign governments, and certainly not our own. If a person cannot be arrested, charged, tried, and convicted through a transparent legal process, they should not be killed. Full stop. There is no exception in the Bill of Rights for “but the computer said so.”

Furthermore, the concentration of AI weapons development in a handful of private companies — funded by the Pentagon, insulated from competition by classified contracts, and subject to minimal oversight — is the military-industrial complex on steroids. Eisenhower warned of the “unwarranted influence” of the defense industry. He could not have imagined a world where private companies build autonomous killing machines that governments deploy with no accountability to anyone.