Graylark : Raven localise un logement sur photo d'intérieur
TL;DR
- Miami-Dade confirms Raven use has been 'limited and largely exploratory' with no significant breakthroughs or arrests, undercutting the product's stated operational value.
- Raven covers three capability modes beyond interior photos: geoestimation, street-level targeting, and vehicle identification, widening the surveillance surface past the single headline use case.
- The law-enforcement pivot predated the 2025 stalking controversy, indicating Graylark repositioned GeoSpy for government clients on commercial grounds before public pressure.
Graylark Technologies commercialise Raven, un outil d'IA qui identifie l'adresse d'un logement à partir d'une seule photo d'intérieur, en recoupant les détails visibles avec les annonces publiées sur Airbnb, Zillow et Vrbo, rapporte 404 Media. Le produit, anciennement nommé GeoSpy, s'appuie sur des modèles entraînés ville par ville plutôt que sur une recherche d'image inversée.
Daniel Heinen, PDG de Graylark, raconte avoir essayé lui-même : « I was curious what it would say and I dropped it in there and it put a pin on San Francisco. » Un cliché aussi banal qu'une machine à laver, sans métadonnées GPS, suffirait à produire une adresse candidate.
Deux clients publics sont nommément cités : le Los Angeles Police Department et le Miami-Dade Sheriff's Office. Interrogé par 404 Media, le bureau de Miami tempère pourtant : « use of GeoSpy has been limited and largely exploratory... has not resulted in any significant investigative breakthroughs. » Après plusieurs mois d'usage, aucune percée d'enquête, aucune arrestation revendiquée.
Graylark a restreint l'accès à son produit en janvier 2025, le réservant aux agences gouvernementales et aux clients entreprises après qu'une version grand public ait circulé. Airbnb, principale source de données exploitée, répond de manière laconique : « Protecting our community's privacy is important to us. » La plateforme n'indique pas si elle entend bloquer techniquement la collecte de ses annonces.
Côté actionnariat, Alexandr Wang, aujourd'hui responsable de l'IA chez Meta, figure parmi les business angels de Graylark. Le dossier rejoint notre fil surveillance, déjà dense cette semaine.
Ce qu'en disent les autres médias
-
International Business Times Lire →
Leads with Miami-Dade's own admission of no investigative breakthroughs; questions the accuracy gap between controlled demo conditions and real casework.
To date, use of GeoSpy has been limited and largely exploratory.
-
Founderland Lire →
Frames the pivot as deliberate product strategy; examines purpose-built government additions including case management and CCTV processing that post-date the public stalking backlash.
The deliberate narrowing from GeoSpy's earlier openness to the OSINT community is complete.
-
Glitchwire Lire →
Details Raven's three operational modes and vision-transformer architecture, broadening the technical picture beyond interior-photo matching to street and vehicle targeting.
The system's vision transformer breaks down a single frame into components like vegetation, soil, architecture, and sky.
-
Reclaim The Net Lire →
Extends the privacy implications to hotel rooms specifically, framing Raven as the end of location anonymity in spaces that previously offered it by structural default.
One of the last places where taking a photo wouldn't give away your location was a hotel room. Now even that is being taken away.
Shared on Bluesky by 5 AI experts
-
Probably nothing to worry about for homeowners unless there are a bunch of other devices taking images in your home and making that data available for sale. Good thing no one would do that.
View on Bluesky → -
New: an AI company now lets cops identify specific houses, using data from Zillow and Airbnb, based on a single photo. It's part of Raven, previously called Geospy. Company previously demoed its tools to catch people van…
View on Bluesky →
Article original publié par 404media.co
Lire l'article original →Titre original : Graylark : son IA Raven localise un logement à partir d'une seule photo d'intérieur, LAPD et Miami en clients