Lethal Beta
Every war is now a product demo. Gaza and Ukraine were the test run. Now Iran is the launch event.
In the tech industry, a product released to real users before it’s fully finished is referred to as being in beta. You collect data on what breaks, who breaks it, and how. You revise. You ship the next version.
For decades, Israel has been running a version of this process on the people of Gaza. The product is weaponry. The beta testers didn’t sign up. And the data collected — on what kills efficiently, what survives countermeasures, what scales — gets packaged and sold.
I’m calling this lethal beta: the systematic use of active conflict zones as live testing environments for AI-powered weapons systems, with the understanding that the resulting “battle-tested” label is the product’s primary marketing asset.
This isn’t a fringe argument. Investigative journalists have documented it, and in some cases its perpetrators have openly admitted it.
The Factory Floor
During Israel’s 11-day conflict in Gaza in May of 2021, The Times of Israel quoted a senior Israeli defense official who proudly called it “the first AI war.”
“For the first time, artificial intelligence represented a key factor and force-multiplier in warfare against an enemy,” the senior intelligence official told the reporter. But it was another three years before we knew what he meant.
In April 2024, the Israeli investigative outlet +972 Magazine published what may be the most important piece of journalism about AI warfare ever written. Based on interviews with six Israeli intelligence officers, it revealed a system called Lavender — an AI tool that had assigned kill ratings to as many as 37,000 Palestinians, ranking each person’s probability of being a Hamas operative on a scale from 1 to 100. Human officers were authorized to approve strikes based on Lavender’s output. The average time spent reviewing each target before authorizing a bombing: about 20 seconds — just long enough to confirm little more than that the target was male. +972 Magazine
A second system, called “Where’s Daddy?”, tracked Lavender’s targets and alerted operators when the intended strike target returned home at night — deliberately timing strikes for when entire families would be present.
A third, “The Gospel,” identified buildings and infrastructure for destruction. Before these AI systems, Israeli analysts sifted through intelligence to verify (or “produce”) around 50 targets per year in Gaza. After implementation, the IDF was reportedly striking up to 250 targets per day. JINSA
The companies making these systems tend to point out that the technology can serve to make strikes more surgical. But they also make them more frictionless, and thus easier to approve. This is not just AI as a tool to improve precision. This is AI as an engine for volume.
Market Fit
There’s a thriving market at the end of the pipeline.
Each time Israel uses new technology in war, it sees a spike in global demand. The latest Gaza war is also the latest laboratory for its arms industry. Al Jazeera’s Antony Loewenstein, whose book The Palestine Laboratory is the definitive text on this dynamic, estimates that Israel has exported hi-tech surveillance tools to at least 130 countries over the last 50 years — to democracies and dictatorships alike — and that one of the main reasons so many nations tolerate Israeli policy is precisely because it allows them to buy these battle-tested weapons. Anadolu Ajansı
In reporting for The Pulitzer Center, Matt Kennard interviewed Shlomo Brom, a retired Israeli brigadier general who had since become an arms industry expert:
I asked Brom if it’s true that Israeli arms companies use the fact that their products have been tested on Palestinians to gain international business. “Of course,” he replied. “Why not? Marketing [professionals] try to use any advantage and if they can use the advantage that this system was tested operationally and it worked, they will of course use it for marketing.”
The revenue figures of the companies show that the strategy Brom describes is good business.
