AI and X-Rays Reveal Hidden Text in 2,000-Year-Old Herculaneum Scrolls

X-Ray Mapping and AI Unlock Charred Herculaneum Scrolls

Published Sept. 16, 2026 in PLOS One, new research demonstrates that noninvasive X-ray imaging combined with artificial intelligence can peer through charred papyrus scrolls from Herculaneum.

For classical scholarship, the breakthrough targets a stubborn wall. When an 18th-century seaside villa yielded roughly 2,000 brittle, blackened rolls, researchers faced an analytical dead end. The carbon-based ink and the carbonized papyrus have similar elemental signatures, making conventional imaging ineffective.

Testing Leaded Inks Against Vesuvius Heat

Seeking a way past the ancient disaster, investigators—including University of California, Berkeley research affiliate Douglas Seiler—created synthetic sample scrolls using leaded ink and charred them to replicate the conditions of Mount Vesuvius’s A.D. 79 eruption.

The workflow relies on a three-step process. First, scientists apply X-ray fluorescence, or XRF, to map elemental composition across the scrolls and confirm the presence of lead in some inks. Next, X-ray tomography constructs high-resolution, three-dimensional images of the rolled material.

Finally, custom machine-learning algorithms analyze the 3D scans. They detect subtle contrasts between carbonized papyrus and lead-containing marks.

Reconstructing Lost Letters at Low Concentrations

“It’s amazing what you can get electrons to do,” said Douglas Seiler, describing how element mapping and imaging can reveal otherwise invisible marks.

In rigorous laboratory tests, the integrated approach successfully detected and reconstructed letters written with lead-bearing inks, even when lead concentrations were very low. Although certain characters remained hidden, the technique achieved legible contrast in scenarios where standard X-ray methods fail.

Prioritizing Fragile Artifacts for Future Study

The Herculaneum collection includes philosophical works attributed to Epicurean and Stoic authors and other classical texts. While past digital unrolling efforts have salvaged isolated passages, large-scale reading remains sporadic because of low contrast against the blackened substrate.

Through the integration of artificial intelligence, three-dimensional X-ray scans, and elemental mapping, scientists have acquired a safe mechanism to identify and examine scrolls featuring lead-based writing—thereby raising the probability of retrieving forgotten literature while leaving delicate manuscripts unopened.

Consequently, the study authors advise conservators to screen original scrolls using XRF for lead signatures prior to detailed tomography and AI-driven reconstruction. Such a pipeline assists in selecting the best targets for analysis while cutting down on manual manipulation and potential harm to invaluable museum pieces.