Reuters Institute (Oxford)

Investigating Atrocities Using AI

A team of journalists entered Damascus after Bashar al‑Assad’s regime fell, digitizing tens of thousands of security‑ministry documents, many handwritten in Arabic, to investigate war crimes. Using a custom AI pipeline, lead technologist Allison Martell built translation layers and a visual‑search model that could parse the images, extract text, and flag signatures. The system sifted through massive noise to surface concrete evidence, such as a distinctive stamp belonging to Colonel Ismander, identified as the regime’s “master of cleansing.” The model’s detection linked Ismander to a scheme that exhumed bodies from one mass grave and reburied them elsewhere, a tactic designed to obscure evidence. The team cites the stamp‑matching as a “signal among the noise,” enabling them to publish multiple, data‑driven stories on atrocities that would have been impossible to assemble manually. By turning unstructured, multilingual archives into searchable intelligence, the AI tools dramatically shortened investigative timelines, bolstering accountability and setting a template for future conflict‑zone reporting.