TL;DR
This stretch of news marks the moment heritage decipherment got a workflow: cheap handheld screening, portable CT scanners that travel to the collection, and machine-learning unrolling — a pipeline pattern that turns heroic one-off scans into something scalable. Meanwhile, the money story is diversifying beyond the private-equity roll-ups seen last cycle: self-funded "gentleman scientists" and anniversary-driven federal humanities grants are both bankrolling real breakthroughs, even as an Indigenous data-sovereignty backlash starts to harden into procurement requirements.
Decipherment becomes a pipeline, not a stunt
The Herculaneum and cuneiform toolkits are cheapening and going portable, shifting the field from unique-artifact heroics toward a repeatable screening-to-read workflow.
"With lead in the ink, you would get a huge friggin' signature, so you really need to be looking for scrolls with lead in them… This is the Holy Grail." — ai-text-decipherment-aeneas-vesuvius
Two peer-reviewed results landed within days of each other. A team led by Douglas Seiler — with UC Berkeley chemists, papyrologist Leah Packard-Grams, and NIST's Center for Neutron Research — carbonized replica scrolls written with lead-spiked ink and showed that lead absorbs up to 25 times the X-rays of charred papyrus, meaning an inexpensive handheld X-ray fluorescence scanner can triage which scrolls are worth expensive CT time (Berkeley News). Separately, a team including assyriologist Cécile Michel and physicist Christian G. Schroer built ENCI, "a transportable high-definition CT scanner… designed for on-site use in museums and archives," and read a ~4,000-year-old sealed letter from Kültepe without breaking its clay envelope (npj Heritage Science).
The pattern matters because the hardware now travels to the collection instead of the reverse — cheap XRF triage, targeted CT, ML virtual unrolling. That's the difference between reading dozens of objects and thousands, and it's also the shape of a hardware-and-services market (see the business section below). Berkeley's recap also confirms Brent Seales' team decoded scroll PHerc. 1667 in 2026, apparently a Stoic commentary, while the $1M second-scroll Vesuvius Challenge grand prize remains unclaimed ai-text-decipherment-aeneas-vesuvius.
What to watch: Whether Italian authorities permit systematic XRF screening of the Herculaneum collection — the number of scrolls that test positive for leaded ink determines whether this pipeline scales at all.
Remote sensing finds the invisible — and the field asks what it's for
Drone LiDAR keeps revealing "hidden" infrastructure beside the world's most-studied sites, even as a new review warns that sensors aren't the same thing as progress.
"One researcher had taken around three hours to travel just 250 metres through the forest using a machete." — non-invasive-subsurface-archaeology-gpr-lidar
A Polish-Peruvian team led by Prof. Mariusz Ziółkowski of the University of Warsaw, with drone-LiDAR work by Dr Bartłomiej Ćmielewski and Peru's Apu Space, reported more than 2.5 km of previously unknown pre-Columbian roads plus undocumented terraces and circular structures on Mount Yanantin, across the river from Machu Picchu — and Ziółkowski expects the eventual total "could be at least twice as large" (HeritageDaily). Notably, precise locations are being kept confidential to protect the remains — site secrecy is now standard operating procedure. Meanwhile, M.B. Rajani's new review in the Journal of the Indian Society of Remote Sensing argues bluntly that "technological progress and disciplinary progress do not necessarily advance in parallel," and reorganizes the field around archaeological questions rather than sensor types (Springer).
The tension is the story: discovery capacity is outrunning analysis capacity, which is exactly where commercial tooling — automated feature extraction, web-GIS platforms, drone-LiDAR services — plugs in. The review is also a useful map of players beyond the usual US/European centers: NIAS Bengaluru, Wrocław, Warsaw's Andean Studies center, Apu Space non-invasive-subsurface-archaeology-gpr-lidar.
What to watch: Whether analysis tooling (point-cloud ML, multi-temporal fusion) closes the gap with discovery rate — or whether the backlog of unprocessed scans becomes the field's defining bottleneck.
The trust backlash goes institutional
Critique of AI in archaeology is moving from journal essays toward concrete funder and repository requirements, while readers now police legitimate discovery writing for "AI smell."
"A model can absorb Indigenous knowledge as training data with no consent, no acknowledgement, no compensation and no community control over how it is used next." — ai-archaeology-ethics-open-access-slow-data
Dr Michelle Richards, Prof Michael-Shawn Fletcher, and Anthony Romano, writing in Pursuit (University of Melbourne), document GenAI fabrications — image searches for "cultural heritage Australia" returning fake giant stone heads — and make prescriptions aimed at institutions: journals should "require evidence of community permission before publishing Indigenous data," and funders should build ICIP into ethics review (Pursuit). The trust problem is bidirectional: a 191-point Hacker News thread on the genuinely newsworthy Saqqara tomb discovery collapsed into a 58-comment argument over whether one paragraph "smells like AI writing" (HN).
This matters commercially because consent and provenance requirements are becoming a procurement gate — any vendor or CRM operation touching Indigenous data will need compliance tooling ai-archaeology-ethics-open-access-slow-data.
What to watch: Whether Australian, EU, or UN bodies convert ICIP recommendations into formal policy — that's the moment "slow data" ethics becomes grant-eligibility language.
The funding map is wider than private equity
Beyond the PE roll-ups tracked last cycle, three new funding lanes surfaced: federal anniversary money, individual philanthropy, and quiet federal rulemaking.
"One might be tempted to imagine the 'gentleman scientist' as a bygone era, but it's alive and well here in Berkeley." — commercial-opportunities-funding-crm-archaeology
Wake Forest won a three-year, $750,000 NEH grant — via the "Rediscovering Our Revolutionary Tradition" program — to rescue thousands of 18th-century Moravian artifacts moldering in "a flood-prone building with mold issues" at Historic Bethabara Park (Wake Forest News); the America's-250th anniversary is an active federal driver, and collections backlog cataloging is a repeatable, fundable service line. On the regulatory side, a ten-day scan of the federal record shows no Section 106 rulemaking — but S.2498, the National Park System Long-Term Lease Investment Act, hit the Senate Calendar, signaling appetite for longer-horizon private investment in park operations (congress.gov).
The business read: three concrete lanes this cycle — portable imaging hardware and scanning-as-a-service for museums (the ENCI model), cheap triage-plus-ML sold as workflow, and AI-assisted collections cataloging contracts riding federal money commercial-opportunities-funding-crm-archaeology.
What to watch: Whether the OIRA/Section 106 final-rule watch breaks its silence — the pending rule is the single biggest variable for CRM compliance timelines.
What surprised us
- The biggest decipherment paper of the week was self-funded. Seiler, who "previously made a living in real estate and banking," paid for the replica-scroll research himself — high-risk heritage-AI R&D doesn't wait for grants when a motivated individual can afford papyrus, a furnace, and NIST beam time commercial-opportunities-funding-crm-archaeology
.
- Battery software read the scrolls. NIST's virtual-unrolling code was adapted from lithium-ion battery "jelly roll" inspection software — cross-domain software reuse, not purpose-built heritage tooling, delivered the key step ai-text-decipherment-aeneas-vesuvius
.
- A real discovery got buried by its prose. The Saqqara tomb story — a 4,400-year-old judge's tomb with pigments intact — was nearly lost under a meta-argument about whether the write-up was AI-assisted. Reader trust is now a first-order problem for legitimate outlets ai-archaeology-ethics-open-access-slow-data
.