An unopened scroll from Herculaneum has been read in full for the first time in history, using machine learning and imaging technology. The physical document remains sealed, untouched, and intact—a breakthrough that sidesteps the irreversible damage that destroyed many scrolls when researchers tried to unroll them centuries ago.

The milestone emerges from the Vesuvius Challenge, an open research competition designed to solve a specific archaeological puzzle: recovering text from carbonized papyrus scrolls buried when Mount Vesuvius erupted in 79 AD. Herculaneum's library—the only intact library from antiquity—was sealed by that eruption and remained largely illegible despite the scrolls' physical recovery. Unopened, they are fragile carbon cylinders with nearly invisible writing.

The Vesuvius Challenge awarded its 2023 Grand Prize to a team that extracted readable passages from an unopened scroll—15 passages containing more than 2,000 characters decoded in February 2024. That result validated the overall approach. The full-scroll read announced subsequently extends the method to completion: the entire document from start to finish, preserved.

How the Pipeline Works

The technique chains together well-established components in a new configuration. Micro-CT scanning—a type of X-ray imaging used in industrial quality control—captures detailed 3D density data from a sealed scroll at sub-millimeter resolution. Researchers then computationally "unwrap" the scroll's layers, reconstructing the geometry of a rolled manuscript without physical contact. This produces a flattened virtual surface that can be examined for traces of ink.

Machine learning becomes essential at the ink-detection stage. Ancient carbon-based ink on carbonized papyrus creates almost no visible contrast in standard CT scans—both the paper and the writing are carbon. Challenge teams trained neural networks to recognize the microscopic texture patterns that ink leaves on papyrus, a signal too subtle for conventional image-processing techniques. Refined across multiple competition rounds, the models learned to extract individual characters from noise that human annotators could not reliably identify on their own.

Luke Farritor, a 21-year-old competitor, became the first person in roughly two thousand years to read a word from inside an unopened scroll. This intermediate milestone validated the ink-detection method before the full pipeline was ready for complete documents. His result fed directly into the broader research effort.

What This Opens

The Herculaneum library is thought to hold hundreds of scrolls, the vast majority still unread. Most surviving ancient philosophical and literary work—especially Epicurean philosophy, the library's probable focus—exists only in fragments or through later copies that accumulated errors over time. A non-invasive pipeline for reading sealed scrolls changes what researchers might reasonably recover.

The passages decoded in February 2024 were attributed to the philosopher Philodemus and addressed the topic of pleasure. By the standard of known Philodemus texts, the content is unremarkable, but that is almost beside the point. The significance rests on method, not subject matter. A pipeline that works on Philodemus works on the scroll adjacent to it, and the one after that.

There is, however, a substantial gap between a proof of concept and a production system. The micro-CT scanning required is time-consuming and generates enormous datasets. The machine-learning models have been trained on a relatively small collection of validated scroll segments. The unwrapping step still demands significant human verification and adjustment. Scaling from one complete scroll to hundreds remains an engineering and resource challenge that researchers have not yet addressed. The first full read shows what is possible; it is not yet a routine process.

What the technical trajectory shows is consistent improvement with each competition iteration: better models, better algorithms, lower error rates. The kind of incremental, competitive, open refinement that produced this result is the structure most likely to accelerate timelines. From the first word to the first complete scroll took roughly two years. The next milestone may arrive faster than that.

The Herculaneum scrolls have waited nearly two thousand years. The tools to read them, as it happens, are arriving at a pace most classicists did not anticipate.