No, AI Agents Didn’t Discover a Lost Maya Site: What Actually Happened

A Google researcher said 48 AI agents found an unmapped Maya site in Mexico. It had been published in 2025. Here is the story and what archaeologists said.

Reviewed by Mindloop Editorial on . Every link in the Sources section at the end was re-checked that day.

Graphic of AI agent chat bubbles over a lidar map of jungle terrain, illustrating the disputed Maya site discovery claim.

Picture this: 48 AI agents squint at laser maps of the Mexican jungle, a government agency says it has no record of the spot they flagged, and the researcher behind it writes that “This is by far the best time ever to be alive.” A few days later, his post gets a correction banner that starts with “it turns out.”

This is the story of the Maya “discovery” that was not, why it fooled a lot of smart people for a few days, and what it says about asking AI to find something new.

The 72-hour discovery

Pietro Miotti, a Google researcher, described the project as a personal side project done on holiday. According to his own blog post, the idea hit on 18 October while he was traveling in Peru and reading about Hiram Bingham and Machu Picchu. He asked Gemini, ChatGPT, Claude and Grok what a modern explorer should use, and all four said lidar.

Lidar, in case you missed our explainer on how lidar sees through a jungle canopy, is a laser-based mapping technique that can reveal ground shapes hidden under trees.

Peru, he wrote, was not well covered by public lidar data, so he switched to the Maya region. By his account, the agents turned about 1,200 square kilometers of lidar strips into images, and 48 agents looked through them for pyramids, platforms and plazas. He says no special classifier or custom model was involved.

By the next day, the agents had flagged what he describes as one major village that appeared not to be registered. His description: a leveled plaza, an oval pyramid at least 8 to 12 meters tall, a raised courtyard and a flat-bottomed hollow that might have been a reservoir.

On 20 October he had a report translated into Spanish and sent it to Mexico’s National Institute of Anthropology and History (INAH). On 21 October, INAH replied that “no registered archaeological site was identified at the location given in your report.”

His blog framed that as under 72 hours from a random thought to an official government reply.

The twist: someone had already been there

According to The Verge, the site “had already been discovered and was published in an academic journal last year.” Miotti’s own blog now carries an update dated 11 November: “it turns out the site had already been mapped in Estrada-Belli et al. 2025.” He added that “The agents missed it while going through the literature,” so, as Futurism quotes him, “they rediscovered it rather than discovered it.”

The paper is by Francisco Estrada-Belli, Marcello Canuto, Ivan Šprajc and Juan Carlos Fernandez-Diaz. Crossref lists it in the Journal of Archaeological Science: Reports in October 2025, titled “New regional-scale Classic Maya population density estimates and settlement distribution models through airborne lidar scanning.” Futurism reports that the work covered about 95,000 square kilometers of lidar data and that Estrada-Belli later shared images of the site with archaeologist Sarah Parcak.

Miotti’s update closes with “The rest of the story still holds.” Archaeologists quoted by The Verge see it differently.

Why the cross-check failed

The detail that makes this story useful is why nobody caught it sooner. Miotti told The Verge: “Archaeologists rightly don’t publish the coordinates of sites, so checking whether a site is already known means comparing images against figures in papers, not searching a database.”

In other words, there is no tidy database to query. A site can be in a peer-reviewed paper, known to local people and absent from a government register, all at once.

Sarah Parcak, an archaeology professor at the University of Alabama at Birmingham, wrote on Bluesky that “This is why we have peer review,” and noted that something missing from a government database is not necessarily unknown or unmapped. INAH’s reply said only that the spot was not registered. Ivan Šprajc, who co-wrote the paper, told The Verge that INAH having no official knowledge of a cluster of features “does not necessarily mean that it has not been spotted or even explored by archaeologists.”

Think of it as checking whether a restaurant exists by asking the tax office. A “no record” answer is accurate and still not the answer to your question.

What the experts said

  • Elizabeth Graham (University College London) told The Verge: “Our ‘problem’ is not discovering new sites.” She said the bigger obstacle is money: “there just isn’t the funding to excavate or map sites as there was in the past.” She also noted that local communities are often already well aware of sites in their area, and credited Miotti with decent work mapping mounds, adding that “the interpretation of lidar data can be tricky.”
  • Ivan Šprajc said “just spotting something does not add much to our knowledge,” and that archaeologists have used lidar and similar tools “for quite some time.” He added that he had received an email from “another AI ‘discoverer’” that same day, and worries about being flooded. He also raised the risk that publicizing locations can attract looters.
  • Miotti told The Verge: “This was a personal project that I was excited to share. I didn’t expect it to get this much attention.” He said that as a non-archaeologist he had not realized that unregistered is not the same as unmapped, and that expert knowledge “is needed to verify what the agents find.”

A fair reading

There are two true things here. The agents did what they were asked: they sorted a very large pile of lidar images and surfaced a plausible settlement, and Miotti said archaeologists and PhD students have contacted him to ask how the workflow works. Graham’s comment that the mapping was decent supports that part.

The second true thing is that finding a shape on a map is only the start of discovery. The hard part is knowing whether anyone has already published it, and that part needs specialist knowledge. The Verge puts the wider pattern bluntly: amateur use of AI tools often shifts the burden of establishing what has been found to the experts.

A quick checklist before you share an AI “discovery”

  1. Ask whether “not in the database” really means “unknown.” It rarely does.
  2. Search the papers, not only registries. In archaeology, coordinates are often left out on purpose.
  3. Contact researchers who work in that region before announcing anything.
  4. Wait for independent checks. In this case, archaeologists spotted the problem quickly and said so online.
  5. Credit earlier work first.

What happens next

As of 15 November 2026, Miotti’s post carries its correction, and The Verge’s piece ends on Šprajc’s worry about being flooded with similar claims. If you want to see how people give agents written instructions for big tasks, our guide to agent skills in Claude Code covers it.

The lidar will still be there next week, and so will the jungle. The archaeologists have a head start.

Sources

Links last checked on 15 November 2026.

Written by Mindloop Editorial

The Mindloop Experience editorial team writes clear, carefully sourced explainers and how-tos on AI and technology, software and the business of tech, for anyone who likes to know how things actually work.

Meet the editorial team

Responses

Write a response

More from Mindloop Experience

See all from Mindloop Experience

You can clap up to 50 times.