The Israeli military uploaded 3.5m photos to Amazon’s cloud platform in order to use the US company’s facial recognition technology to analyse images of Palestinians, the Guardian can reveal.
Leaked documents show the expansive catalogue of images was moved into Amazon’s cloud in the early months of Israel’s devastating aerial bombardment and ground offensive in Gaza.
The images were uploaded by a department of the Israeli military that holds a vast collection of photos covering the Palestinian populations in Gaza and the West Bank.
The department informed Amazon of the decision to upload the images in early 2024 so it could use Rekognition, the company’s AI-powered image analysis platform for a facial recognition project.
The documents obtained by the Guardian form part of an investigation into the Israeli military’s reliance on the technology and infrastructure of Amazon’s cloud computing business to monitor Palestinians.
According to the files and interviews with military sources, Israel has in recent years migrated huge volumes of data – including sensitive intelligence material and information about Palestinians living under occupation – into Amazon’s datacentres.
The large set of images was moved into an Amazon datacentre near Tel Aviv by a department commonly known as Rolling Stone, which sits within a military software unit.
At the time, the department was exploring using facial recognition technologies to analyse the extensive catalogues of images of Palestinians it holds, several sources with knowledge of its work said.
After Israel’s ground invasion of Gaza in October 2023, its military routinely collected images of Palestinians as they passed through checkpoints, often after they had been forcibly displaced by the bombing campaign.
It’s not clear whether the 3.5m images uploaded to Amazon’s Rekognition were captured at these checkpoints, which effectively served as pop-up surveillance posts.
However, a military source familiar with facial recognition efforts at the checkpoints said Amazon’s cloud infrastructure was used due to the “very large scale” of images collected at the sites.