One Airbnb interior photo. LAPD gets the exact address.

This isn’t an estimate. On October 12, 404 Media documented that both the Los Angeles Police Department (LAPD) and the Miami-Dade Sheriff’s Office have bought access to Raven, the tool previously called GeoSpy[1][3]. You upload one interior photo — no GPS metadata, no view out the window — and it hands back the street address[2]. The training corpus is consumer platform listings: 404 Media found result photos identical to ones posted on Zillow, and results that quoted Airbnb and Vrbo listings verbatim, in Charlotte and Montecito respectively[1].

The platforms answered. Airbnb said it has no relationship with Graylark and that its terms prohibit scraping; Zillow said scraping violates its terms of use[5]. Graylark has not explained how it obtained the imagery.

What makes this matter to your computer vision stack isn’t the geolocation itself — it’s the architecture. Raven is not the only actor in the chain: there’s a model, and there’s a third-party corpus nobody gave permission for. For any AI/CCTV RFP, that turns into three clauses:

  1. Training-corpus disclosure clause. The vendor must state in writing what data each model was trained on — including whether it used listings, photos, or text from third-party consumer platforms — and whether that use was authorized.
  2. Platform-consent clause. If a model was trained on data from a platform (Zillow, Airbnb, a marketplace), the vendor must be able to demonstrate that platform’s verifiable consent.
  3. Customer-revocable model deployment clause. The customer — you — reserves the right to revoke the use of any model trained on data with undemonstrated consent, with no contractual penalty.

One honest qualifier the coverage leaves out: Raven’s Find Property feature describes itself as beta, one supported city at a time, and its similarity score is, in the vendor’s own words, “not a probability that an address is correct”[4]. The product is not infallible yet. The consent architecture, though, is already settled — and it isn’t.

P.D. Could your video analytics vendor tell you today what data its person-detection model was trained on? If the answer is “I don’t know,” you have your clause.

Sources

  1. 404 Media (Joseph Cox), Oct 12 2026 — https://www.404media.co/ai-company-lets-cops-pinpoint-specific-houses-based-on-a-single-interior-photo/
  2. State of Surveillance, Oct 13 2026 — https://stateofsurveillance.org/news/daily-surveillance-briefing-october-13-2026
  3. Graylark (vendor) — https://graylark.com/geospy
  4. Graylark (vendor) — https://graylark.com/raven/capabilities/find-property
  5. Malwarebytes (Danny Bradbury), Oct 2026 — https://malwarebytes.com/blog/privacy/2026/10/raven-ai-lets-cops-find-your-home-from-a-photo-taken-inside