AI scours 400 years of archives, finds lost wildlife and meteorite
Original: Pointing AI at archives found a forgotten meteorite, lost rhinos, and more
Why This Matters
Shows AI can accelerate archival research at scale while keeping human-verifiable sourcing.
Software engineer Jesse Waites used AI agents to scan 4.35M pages of Dutch East India Company records and centuries of digitized newspapers, surfacing an unrecorded meteorite, three lost rhinos, and unreported volcanic eruptions.
Inspired by historian Benjamin Breen's October 2026 post—in which Breen used Claude Opus 5.5 to find a 1615 eyewitness dodo sighting in Dutch East India Company (VOC) records—software engineer Jesse Waites set out to scale the approach. Where Breen searched manually, Waites built an automated AI pipeline to hunt multiple historical mysteries in parallel.
The corpus was enormous: 4.35 million handwritten VOC pages digitized by the GLOBALISE project, Dutch national library newspapers, two centuries of American papers, and ship logbooks. Waites estimates a human reader would need roughly 70 years to read the VOC archive alone at a reasonable pace. His home AI lab processed it in a single 12-hour overnight run.
The pipeline started with an AI 'Deep Research' assistant ranking 13 candidate mysteries by data availability and verifiability. Every claim had to trace back to a real, publicly accessible archival document—a direct response to the known problem of AI hallucination. Waites and Anthropic's Claude Code then co-designed the workflow, which layered multiple models to filter noise and handle OCR errors. Old Dutch alone spells 'rhinoceros' roughly 15 different ways.
Results included a forgotten meteorite report, three previously unrecorded rhinoceros sightings, and volcanic eruptions missing from the scientific record—including candidates related to the mysterious unlocated 1808 eruption.