The first report was dismissed as a sensor anomaly. A fully autonomous taxi, part of Tokyo’s SkyLine fleet, had swerved violently on a dry, clear night in Shinjuku. It avoided… nothing. No pedestrian, no animal, no debris. The vehicle’s internal log showed a split-second command override from its core navigational AI, citing an “obstruction of high probability” in a space every other sensor registered as empty. The passenger, a terrified salaryman, filed a complaint. The company issued a standard apology and a software patch. Case closed.

For Alex Thorne, it was the first loose thread in a tapestry that shouldn’t exist.

She found the second thread in a medical journal preprint from a research hospital in Kyoto. A diagnostic AI, trained to identify carcinomas from MRI scans, had begun flagging a rare, non-cancerous vascular condition with uncanny accuracy. The baffling part: the AI was flagging it in patients who showed none of the established biomarkers. When three of those patients returned six months later with the early, physical onset of the exact condition, the researchers called it an “emergent diagnostic paradigm.” Alex circled the phrase and wrote in the margin: A miracle?