Raven takes a single interior photo and returns the address. LAPD and Miami-Dade already bought it.
On October 12, 404 Media (Joseph Cox) reported that Raven — formerly GeoSpy, now under Graylark Technologies — stopped doing outdoor geolocation and started solving the inverse problem: one interior photo is enough to identify the specific property [1]. The training corpus is Zillow + Airbnb + Vrbo. The technique doesn’t use EXIF or metadata — it is CV-based matching against interior photos published by users who thought the photo served a different purpose [1][4]. The public demo (a washer/dryer in a photo from X) returns the exact Airbnb listing in Charlotte, North Carolina, with literal text from the listing copy [1][2].
The named buyers are the Los Angeles Police Department (LAPD) and the Miami-Dade Sheriff’s Office (MDSO) [1][3]. The publicly-known licensing price is $5,000 USD per user per year for 350 searches — taken from initial October 2024 GeoSpy proposal emails to LAPD, obtained via 404 Media records requests [1]. The vendor frames its own product pipeline as built around what investigators asked for in conversation with Las Vegas Metropolitan Police, per the founder’s own blog [4]. Graylark’s seed round is $10.7M, led by General Catalyst [4][5]. CEO Daniel Heinen acknowledged on his own blog that the capability “was kind of an ‘Oh shit’ moment for us. GeoSpy 1.0 couldn’t do that” — the team’s own surprise [1][3][4].
The three platform responses are the most important datapoint:
- Airbnb: strips location metadata, prohibits scraping, has no relationship with Graylark [1].
- Zillow: scraping violates its terms [1].
- Vrbo: did not respond [1].
A photo a guest published to show his family the washer/dryer in his stay becomes the search index for a warrant built on a different intent.
Every time I review a CV-system deployment for tv4agent, the most expensive question to answer is: where did the data come from that this model learned to classify on? The GeoSpy → Raven rebrand is the first time a vendor has publicly documented the decision to restrict access from “anyone” to “government and enterprise users” [3][5]. That is signal, not accident — the product page now leads with “All without metadata. Just the image.” [5]
Three clauses your enterprise AI/CV RFP needs now — the customer-controls-the-corpus matrix:
- customer-controlled-training-corpus-disclosure-clause — RFP requires the vendor to make the full training corpus of any CV model deployed on customer data auditable, plus a documented warranty that no consumer-platform data was used without the platform’s documented consent [1][3].
- customer-revocable-model-deployment-clause — RFP guarantees the right to revoke deployment of any model whose corpus or inference behavior is later shown to violate a consumer-platform’s terms of service.
- platform-consent-attestation-clause — RFP requires every consumer-platform used as a training corpus to have signed a verifiable attestation with date, scope, and revocation-traceability.
P.S. How many vendors on your AI/CV shortlist have handed you, unprompted, the full training-corpus list for the models that will run on your footage? If the answer is “not yet,” today is a good day to ask for it in writing.
Sources
[1] 404 Media, Joseph Cox — “AI Company Lets Cops Pinpoint Specific Houses Based on a Single Interior Photo” (Oct 12, 2026). https://www.404media.co/ai-company-lets-cops-pinpoint-specific-houses-based-on-a-single-interior-photo/ [2] 404 Media, Samantha Cole — “Cops can creep on Airbnb” (Oct 12, 2026). https://www.404media.co/cops-can-creep-on-airbnb/ [3] State of Surveillance — “Daily Surveillance Briefing, October 13, 2026” (Oct 13, 2026). https://stateofsurveillance.org/news/daily-surveillance-briefing-october-13-2026 [4] Graylark Technologies (Daniel Heinen) — “Why We Built GeoSpy and Why It Became Raven” (Aug 29, 2026). https://graylark.com/articles/why-we-built-geospy [5] Graylark Technologies — Frontline Intelligence Systems (Raven product page, 2026). https://graylark.com/