Vesuvius scroll AI research has helped scholars do something that once seemed cruelly impossible: read an ancient manuscript while leaving it physically rolled up. The Herculaneum papyri were scorched and buried when Mount Vesuvius erupted in AD 79. Many survive as fragile carbonised rolls that could be destroyed by conventional opening. Scientists and specialists have used detailed X-ray scans, computer reconstruction and machine learning to recover writing from inside the damaged material.
A major milestone was announced in June 2026, when the Vesuvius Challenge said a team had read the surviving text of one sealed scroll continuously from end to end. Its original announcement identifies the manuscript as PHerc. 1667 and explains that it was virtually unwrapped rather than physically opened. This is a remarkable achievement with a limitation worth keeping close: the researchers read the surviving core of a scroll whose outer portions had already been lost in earlier attempts to open it.
How Vesuvius scroll AI works without touching the pages
Ordinary books invite a simple action: open a cover and turn the pages. The Herculaneum papyri cannot safely be treated that way. Their layers are charred, compacted and intertwined, and attempts to separate them physically can destroy the very writing that scholars hope to study. The challenge is therefore not just recognising ancient letters. It is first finding the page inside a damaged three-dimensional object.
Researchers use high-resolution X-ray imaging to examine the interior of a roll. The scan reveals layers winding around each other, but those layers do not lie like neat sheets in a modern notebook. They are bent, crushed and sometimes broken. Computer methods must trace a surface through the scan and reconstruct it as a flatter sheet that can be inspected for writing.
Only then can ink detection become useful. In these manuscripts, ancient ink and carbonised papyrus can be difficult to distinguish in an image. Machine-learning models help identify subtle signals associated with the ink and show where letters may be present. The resulting marks are not automatically a reliable translation. Specialists in ancient writing and language must still examine the forms, check patterns and interpret the text.
Reading a page also requires keeping track of where every mark belongs. If software mistakenly flattens one layer onto another, it might combine unrelated letters into an apparently plausible sequence. The image-reconstruction process and the language-reading process therefore provide different opportunities to catch mistakes. Scholars can ask whether a proposed line fits the physical surface as well as the grammar and context of ancient Greek. That layered scrutiny helps explain why the project makes scan data and methods available for others to inspect.
The process joins very different kinds of expertise. Physicists and imaging specialists contribute the scan, computer scientists work on reconstructing the layers, machine-learning researchers improve the ink signals and papyrologists assess the ancient Greek. Calling the result an AI discovery is convenient, but it would misrepresent the achievement if it suggested that a chatbot simply looked at a charred object and produced a finished book.
What was actually recovered from the sealed manuscript
The June 2026 announcement concerns PHerc. 1667, also known in the project as Scroll 4. The surviving compact inner portion was reconstructed into a readable surface containing around twenty-two columns of Greek. The project says that earlier attempts to open the object in past centuries had removed or destroyed parts of the original roll. The phrase complete reading therefore refers to the text surviving in that rolled core, not to an untouched manuscript preserved from its first to last original page.
The recovered writing appears to be a philosophical treatment of ethics with Stoic associations. The project links parts of the text to themes of human nature, impulse and moral development, while noting damage and gaps in the readable passages. Attribution and interpretation involve scholarly assessment. A technical success in revealing marks is not automatically a final answer to which ancient author wrote every surviving line.
This achievement also drew strength from earlier work. In the 2024 Grand Prize announcement, the Vesuvius Challenge described the first substantial passages recovered from a different scroll, involving ideas about pleasures such as music and food. Those early results were a partial reading, not the same event as the continuous reading reported in June 2026. Keeping the dates and manuscripts distinct shows how the project advanced over time.
The dates of the milestones matter. The major passage-reading prize was announced in February 2024, while the continuous reading of the surviving part of PHerc. 1667 was reported on 25 June 2026. The second result was not merely a reannouncement of the first. It dealt with another scroll and a harder goal of recovering continuous text across the available manuscript. Looking at the sequence avoids both understating earlier work and exaggerating how quickly the latest achievement occurred.
A further development involved improved imaging of ink inside another damaged roll and identification of a title associated with Philodemus in a separate manuscript. These projects illustrate why there is no single moment when the entire Herculaneum library suddenly becomes legible. Each roll presents a different physical and historical problem, and each claimed reading has to be checked on its own merits.
Why archaeologists cannot simply open the scrolls
Conservation often involves resisting the temptation to disturb a fragile object. A scroll may preserve information precisely because it remained closed while buried, even though that condition made the writing unreachable. Mechanical intervention can flatten a surface at the cost of losing material and context. Digital reconstruction offers a way to examine parts of the interior before deciding what, if anything, should be physically handled.
That is not a promise that computer models are perfect. The shapes reconstructed from a scan may be ambiguous, particularly where the layers are crushed together. Software can accidentally join areas that belong to different surfaces, while a faint pattern could be mistaken for a letter. Scholarly verification and open access to underlying data are important precisely because a convincing image can still represent a mistaken reconstruction.
The work demonstrates how archaeology increasingly combines historical expertise with computation. LiveAIWire has previously explored AI-assisted analysis of archaeological sites, digital preservation of cultural heritage and the use of AI to locate ancient drawings in Peru. Virtual unwrapping is distinct from finding a buried site or enlarging a photograph. It tries to recover legible text from a manuscript that is already known but physically inaccessible.
The prize structure also matters. The Vesuvius Challenge invited a distributed group of researchers to work on shared data and problems, then rewarded verifiable progress. This does not eliminate scientific competition or error, but it allows different people to test methods against common material. The outcome is best understood as an example of specialised tools and open collaboration reinforcing the judgement of historians.
What a recovered ancient book can and cannot tell us
The human stakes extend well beyond the novelty of reading an old roll. Ancient philosophical works often survive only through copies, quotations or fragments, leaving scholars to reconstruct arguments from incomplete evidence. Previously inaccessible writing can provide direct material that changes interpretations, supplies missing passages or shows how thinkers expressed ideas in their own words.
Yet a recovered passage does not arrive with a ready-made modern meaning. Ancient Greek words can have several possible readings, damaged lines leave gaps and philosophical context matters. A machine may help expose the strokes on a page, but translating and interpreting them still require debate. Readers should expect revisions as scholars scrutinise transcripts, compare sources and examine difficult sections.
There is also no reason to assume that every unread manuscript contains a sensational secret. Much recovered material may be repetitive, technical, mundane or already partly understood from other evidence. That does not make it unimportant. Sometimes a small clarification about an author, title or argument is the finding that historians value most, even when it cannot be expressed as a world-changing revelation.
The public availability of scans and transcriptions also offers a valuable form of accountability. Other specialists can compare the reconstructed surface with an image, challenge the reading of a particular mark or produce an improved reconstruction. That openness will not make disagreements disappear, but it creates a path for resolving them with evidence. Ancient texts often become clearer through successive editions; digitally recovered works are likely to follow a similar process as methods and scholarly interpretations improve.
The exciting change is methodological. Researchers can now contemplate studying fragile books without destroying the material that protects them. The work remains slow, dependent on specialist equipment and careful interpretation. But a library silenced by an eruption has begun to yield readable pages again, showing how modern imaging and human scholarship can recover voices that physical force would only have erased.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.