Researchers have for the first time virtually unwrapped and read an entire sealed Herculaneum papyrus scroll, carbonized nearly two thousand years ago when Mount Vesuvius buried the ancient Roman city. The scroll, cataloged as PHerc. 1667, was scanned using high-resolution phase-contrast X-ray imaging at the European Synchrotron Radiation Facility and then decoded through machine-learning ink detection and computational unrolling. The result is a complete text recovered without physically opening the fragile artifact, a feat that had eluded scholars for more than two centuries.
Why a fully read sealed scroll changes the stakes for ancient texts
Previous efforts had extracted individual words or short passages from Herculaneum fragments, but PHerc. 1667 represents the first time a rolled papyrus has been digitally unrolled and read from beginning to end. That distinction matters because hundreds of similar scrolls remain sealed in storage at the National Library of Naples. Many are believed to contain lost works by the Epicurean philosopher Philodemus and possibly other ancient authors whose writings survive nowhere else. A reliable, non-destructive reading pipeline could unlock an entire library that physical conservation methods have never been able to access safely.
The technical paper describing the achievement was posted on arXiv and details how the team combined propagation-based phase-contrast synchrotron micro-CT scanning with improved virtual-unrolling algorithms and trained machine-learning models to detect carbon ink on carbonized papyrus. The scanning took place on BM18, a beamline at the European Synchrotron Radiation Facility designed specifically for multiresolution tomography. BM18’s high spatial resolution allowed the team to distinguish ink traces from the surrounding charred fibers at a level of detail that earlier beamlines could not consistently achieve.
If this pipeline proves repeatable across different scroll conditions, the practical consequence is significant. Scholars have long treated the Herculaneum collection as a frozen archive, too fragile to open and too opaque to scan. A working digital reading method turns that archive into an active source of new ancient literature for the first time since the scrolls were excavated in the eighteenth century.
From early X-ray attempts to a complete digital reading of PHerc. 1667
The 2026 result did not appear in isolation. It sits at the end of a research arc stretching back more than a decade. Earlier phase-contrast tomography experiments on Herculaneum papyri, published in Scientific Reports, demonstrated that X-ray imaging could reveal internal structures of rolled scrolls and even detect traces of writing. Those studies established the basic physics but struggled with resolution and the difficulty of separating tightly compressed papyrus layers computationally.
A separate line of work built the data infrastructure needed to train and validate machine-learning models. The EduceLab-Scrolls project created a dataset with ground-truth alignment between X-ray CT scans and infrared photographs of opened Herculaneum fragments. That verifiable pipeline gave researchers a way to measure false positives and reduce hallucinated text, a persistent risk when algorithms attempt to read faint signals in noisy scan data.
Institutional scanning programs also contributed. Diamond Light Source, the United Kingdom’s national synchrotron, hosted high-resolution scans of Herculaneum materials and tested Brent Seales’ virtual unwrapping process on scroll fragments. Those sessions refined acquisition protocols and demonstrated that synchrotron-grade imaging could be applied to cultural heritage objects without damaging them.
The Vesuvius Challenge, a public competition launched in 2023, accelerated progress by offering prizes for reading text inside sealed scrolls. Early contest entrants decoded individual Greek letters and then short passages, milestones covered by Nature in 2023 and 2024. The 2026 paper on PHerc. 1667 goes well beyond those incremental wins by delivering a complete, continuous text from a fully sealed scroll, satisfying explicit coverage and papyrological review criteria laid out in the technical report.
Open questions about scaling, validation, and the remaining scrolls
Several gaps remain between this single success and a systematic reading of the Herculaneum collection. The full transcribed Greek text and detailed papyrological commentary from PHerc. 1667 have not yet been published beyond the technical summary in the arXiv paper. Independent scholars have not had the chance to compare the machine-generated reading against their own analysis, and direct statements from papyrology reviewers are not yet available in public records.
Ground-truth infrared alignments for the complete scroll, the kind used to validate the EduceLab fragment pipeline, are referenced in the paper but have not been released. Without that data, outside researchers cannot fully replicate the ink-detection results or measure error rates across the entire text. Official ESRF BM18 acquisition logs and beamtime metadata have also not been deposited in public institutional repositories, leaving questions about how much scanning time and parameter tuning were required to reach readable quality.
Scaling the method to the hundreds of remaining scrolls will also demand careful triage. Not all papyri are equally well preserved; some are crushed, folded, or fused into irregular lumps. The algorithms that worked on PHerc. 1667 may need substantial adjustment to handle different internal geometries and damage patterns. In addition, synchrotron beamtime is scarce and competitive. Even if each scroll can be scanned in a matter of hours, scheduling and funding a campaign large enough to cover the entire Herculaneum collection will require coordination among libraries, beamline scientists, and cultural heritage agencies.
Validation will remain a central concern. Machine-learning systems trained on partial datasets can overfit to familiar letter shapes or textures, seeing words where none exist. The fragment-based benchmarks developed for the EduceLab project help constrain that risk, but they do not automatically transfer to full, densely written rolls. To maintain scholarly confidence, future releases will likely need side-by-side presentations of raw volumetric slices, ink-probability maps, and proposed transcriptions, allowing independent experts to check that every letter is supported by visible signal rather than algorithmic guesswork.
What PHerc. 1667 suggests about the future of digital papyrology
Despite those caveats, the successful reading of PHerc. 1667 marks a turning point. It shows that non-invasive imaging and computation can recover long, coherent texts from artifacts that would disintegrate if physically opened. For papyrology, that means research agendas can shift from hypothetical reconstructions of lost works to direct engagement with newly accessible primary sources. For the broader study of antiquity, it raises the possibility that familiar narratives about philosophy, literature, and daily life in the Roman world will be revised in light of texts that have been silent since antiquity.
The project also illustrates a model for cross-disciplinary collaboration. Physicists optimized beamline settings, computer scientists designed unrolling and ink-detection pipelines, and classicists defined what counts as a readable, responsibly edited text. Public engagement, through initiatives like the Vesuvius Challenge, added an open-science dimension, bringing external coders and analysts into a domain once limited to a handful of specialists.
Looking ahead, the techniques proven on PHerc. 1667 could extend beyond Herculaneum. Other carbonized or tightly bound manuscripts-from Dead Sea Scroll fragments to medieval codices with inaccessible bindings-present similar challenges of fragility and opacity. Phase-contrast imaging and virtual unwrapping will not be a universal solution, but they now stand as practical tools that cultural institutions can consider alongside more traditional conservation methods.
For now, PHerc. 1667 serves as both a demonstration and a test case. Its successful reading confirms that the combination of high-resolution synchrotron scanning, sophisticated unrolling software, and rigorously trained machine-learning models can transform an inert black cylinder into a legible book. The next steps-releasing full data, inviting independent verification, and methodically tackling additional scrolls-will determine whether this breakthrough remains a singular achievement or becomes the foundation of a new era in recovering the written record of the ancient world.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
