---
title: "Your Airbnb listing photos are also a map, and police can now read it"
description: "Graylark’s Raven, formerly GeoSpy, claims to match a single interior photo to a specific rental or home, apparently drawing on Airbnb and Zillow"
author: "Nate Ledger"
published: 2026-10-12T13:00:05Z
modified: 2026-10-12T15:58:46Z
url: https://rews.cc/a/your-airbnb-listing-photos-are-also-a-map-and-police-can-now-067ce4
language: en
tags: ["ai", "surveillance", "privacy", "crime", "drones", "tech"]
publisher: "Rews (https://rews.cc)"
---

# Your Airbnb listing photos are also a map, and police can now read it

*Graylark’s Raven, formerly GeoSpy, claims to match a single interior photo to a specific rental or home, apparently drawing on Airbnb and Zillow*

By Nate Ledger · October 12, 2026 · https://rews.cc/a/your-airbnb-listing-photos-are-also-a-map-and-police-can-now-067ce4

## In brief

- Graylark Technologies’ Raven, formerly GeoSpy, now claims to match interior photos to specific rental properties and homes
- Demo results quoted Airbnb and Vrbo listing titles; one demo returned photos identical to those posted on Zillow
- Airbnb and Zillow say scraping violates their terms; Airbnb says it has no relationship with Graylark and is reviewing the case
- The LAPD and Miami-Dade Sheriff’s Office bought access; Miami-Dade says its use has produced no significant breakthroughs or arrests
- Graylark limited access to government and enterprise users in 2025 after Discord users sought help stalking women

Here is something that now apparently exists. A police officer has a photo of a room, say a bathroom or a laundry nook, with no location data attached. The officer uploads it to a piece of software called Raven, picks a region to search, and a couple of seconds later the map zooms in on one particular property and shows the title of its rental listing. [Joseph Cox reported for 404 Media](https://404media.co/ai-company-lets-cops-pinpoint-specific-houses-based-on-a-single-interior-photo/) that Raven, which is made by Graylark Technologies and used to be called GeoSpy, has added the ability to analyze the insides of homes, and that it is clearly pulling data from sites like Zillow and Airbnb.

Earlier this month Graylark’s chief executive, Daniel Heinen, announced on LinkedIn that the company was introducing “the ability to geolocate interior images across millions of short-term rental units worldwide.” The demo video goes like this. Someone takes a photo of a washer/dryer unit from X, uploads it, selects “Find Property: Interior and exterior property geolocation,” and chooses a region. The map lands on Charlotte, North Carolina, and the top result reads “Renovated! Walk Uptown/located in 2 Hot Spots!” That is the wording of a specific Charlotte Airbnb listing, and that listing’s photos show the same property. Later in the video, a bathroom photo from social media resolves to Montecito, California, with the top result “Montecito Miramar Beach House on the Sand,” which 404 Media found is the text of a listing on Vrbo. A demo from last year traces an interior photo to a specific San Francisco address and returns other interior photos of the place; they are exactly the photos posted on Zillow. Another video, posted last week, identifies a particular hotel.

![A red location pin contains a laundry room with two washing machines, shelves, and plants, pointing to a cream house icon on a blue background.](https://rews.cc/img/185d135e38b2b7be863f2d5cb3ca9ee9849c3732.webp)

It is worth being precise about what this is, because it is a different trick from the one you might assume. A reverse image search takes your picture and finds the same picture somewhere else online. Raven instead seems to pick out the distinctive features of a room and then find other photos of that room, which can be different shots taken at different times. You don’t need the identical image to exist anywhere. You just need the room to have been photographed by someone, somewhere, with a label attached.

Which brings us to the labels.

Heinen’s posts don’t say where the property data comes from. Airbnb told 404 Media that “location metadata is excluded from listing photos, hosts are given the option to show only an approximate location of their listing, and a home’s address is shared with a guest only after a reservation is confirmed,” and that it has “technical safeguards in place to try to limit automated data collection from our site.” It said it has no relationship with Graylark, that its terms of service prohibit scraping, and that it is reviewing the case. Zillow said scraping breaks its terms of use and that it pursues legal avenues when necessary. Vrbo did not respond to multiple requests for comment.

I think the interesting thing here is how a rental listing actually works. Airbnb’s privacy setup is: keep the address back, show the photos. But the photos are the whole sales pitch. A host has every reason to shoot the room as clearly and distinctively as possible, with the quirky tiles and the nice washer/dryer, because that’s what gets people to book. Then the platform attaches a title and an approximate neighbourhood. From a machine-learning point of view, that is a labelled dataset that millions of hosts built for free, each one trying hard to make their room unmistakable. The address is hidden. The thing that gives away the address is the product.

![A coral sofa, round coffee table, potted plant, and curved floor lamp sit beneath cream, coral, and navy arches on a bright blue background.](https://rews.cc/img/16e5cb2be87a045a05acaa39bbce70168427b06a.webp)

Graylark didn’t set out to build this, by its founder’s account. [Speaking to the tech publication Indicator](https://indicator.media/p/ai-enabled-geolocation-a-conversation-with-geospy-daniel-heinen) last year, Heinen described how the indoor capability showed up during testing of the second version of the model:

> I’ll be honest, we didn’t train it to do that. That was kind of an ‘Oh shit’ moment for us. GeoSpy 1.0 couldn’t do that. As we trained GeoSpy 2.0, we were just kind of red teaming and testing it out and I think I had a picture of my girlfriend just taking a selfie in the bathroom. I was curious what it would say and I dropped it in there and it put a pin on San Francisco.

He then went to Zillow, screenshotted images, and “nine out of 10 of them it would get the right city.” His explanation was that the training data, geotagged images from all over that the team had assumed were outdoor shots, probably included some indoor ones, “whether it’s people at a party or just taking pictures of food on their counter. The models must have learned some details that relate to different locations.” The tool was originally built to read outdoor clues like architecture and soil. The leap from “right city” to “right house” seems to come from narrower models: one part of the interface reads “City-specific property AI to determine an address,” and emails 404 Media obtained earlier showed the Miami-Dade Sheriff’s Office had access to a custom model trained specifically for Miami-Dade County. Heinen’s roadmap: “In 2026, we’re investing heavily in interior geolocation across U.S. hotels, short-term rentals, houses, and apartments.” Houses and apartments get listing photos too, whenever they go up for sale or rent.

![A cream-colored key contains a miniature bathroom with a bathtub, sink, mirror, towel, and red-and-cream checkered floor against a bright blue background.](https://rews.cc/img/578ec536aa39edac62bc99d7e99b3ee63b26786f.webp)

The pitch to police is serious. “For investigators working trafficking and exploitation cases,” Heinen wrote, “that connection can turn an unidentified room into a location lead they can verify and pursue.” The Los Angeles Police Department and the Miami-Dade Sheriff’s Office have both bought access. How well it works in real cases is less clear. Joseph R. Peguero Rivera of the sheriff’s office public affairs team previously told 404 Media that use of the tool “has been limited and largely exploratory,” that it “has not resulted in any significant investigative breakthroughs or arrests,” and that “No case outcomes were driven solely or primarily by information generated by the tool.” Nobody outside the company knows whether the marketing demos reflect what Raven can normally do, or how fast.

The history explains why people worry about who gets to use it. When 404 Media first wrote about the tool in January 2025, essentially anyone could use it, and some people in the company’s Discord server asked for help stalking women. A day after 404 Media asked Graylark for comment, the company limited access to government agencies and enterprise users. GeoSpy has since [become Raven](https://graylark.com/geospy), with a [launch announcement](https://softwarenewswire.com/press-release/graylark-technologies-launches-raven-the-next-evolution-of-geospy-302768464-192211/) presenting it as a broader visual-intelligence platform. In a September post on X, Heinen said Alexandr Wang, now head of AI at Meta, was an angel investor in Graylark; Wang did not respond to a request for comment. And earlier this year Heinen showed Raven being run on a video of people vandalizing a Flock camera, which is a neat example of surveillance products protecting each other. They are also starting to plug into each other: [another vendor recently pitched](https://rews.cc/a/vidizmo-pitches-police-facial-recognition-using-flock-camera-cb6aa5) facial recognition built on Flock camera data, part of a broader [spread of identification tools](https://rews.cc/a/kashmir-hill-says-facial-recognition-spread-to-glasses-apps-84f766) into everyday products.

Airbnb won’t give you the address until you book. To get people to book, it shows them photos of the place. To a well-trained model, it turns out, those were never really two different pieces of information.