Some books are too fragile to open. That makes reading them a very different kind of problem.
On June 25, 2026, the Vesuvius Challenge team announced that it had virtually unwrapped and read the surviving text of a carbonized Herculaneum scroll, PHerc. 1667, from end to end. The scroll was preserved by the eruption of Mount Vesuvius in AD 79, but earlier physical attempts to open it had destroyed outer layers. The achievement concerns what remained, not a perfectly restored original book. Read the team's announcement.
Different people, different parts of the job
High-resolution X-ray scans captured the rolled material. Computational methods reconstructed and flattened its surface. Machine-learning models helped reveal ink against the darkened papyrus. Specialists in ancient papyrus texts then examined and transcribed the writing.
The researchers describe the methods in a June 27 preprint. That is a research manuscript, not a claim here of completed journal peer review. The damaged text still has gaps. Read the research.
What makes the collaboration interesting
Our takeaway is the care in how the work was divided. Better images, computational tools, and specialist reading each addressed a different obstacle. The physical object and the evidence remained central.
That is a useful pattern to notice in smaller projects, too. Before bringing AI into a task, ask: which part is actually difficult? Finding material? Organizing it? Interpreting it? Checking whether the result makes sense?
Those questions can lead to very different choices about where a tool belongs and where a person needs to stay involved.
Most of us will not be reading an ancient scroll this week. We can still borrow the habit of naming each step, keeping uncertainty visible, and checking the evidence before celebrating the result.
