Breadcrumbs

Privacy tools check one photo at a time. A stranger looks at all of them. A park, a trail and a street downtown say nothing alone, but together they can point to where you live, work or walk. Breadcrumbs is a privacy red team for your own photos, built on Meta's Muse Spark and SAM 3.1. 1. Check each post. Muse Spark looks for signs and landmarks in every photo on its own, the way most checkers do. 2. Find the tipping point. An attacker agent reads your feed in posting order (post 1, then posts 1-2, and so on), using tool calls: Spark's hosted web search, OpenStreetMap lookups and distance checks. It marks the exact post where you became findable. On the staged sample, Spark names no place in any of the 5 photos, yet the attacker finds Downtown Columbia, Missouri from posts 1-3 together. 3. Break the pattern. A defender agent reads the attacker's reasoning and makes the fewest changes. SAM 3.1 blurs every sign, letter and logo, and Muse Spark checks again that the place can no longer be read. If the blur does not hold, the post is hidden. The agents repeat until the attacker fails, and the app reports the privacy cost. 4. Check before you post. A browser extension runs the check inside a post composer the moment you pick a photo, and can swap in a blurred version. You can upload photos or import your own Instagram or Facebook data export. Breadcrumbs works only on your own photos, never identifies people or homes, and says plainly that a model failing to find you is not safety from people. The sample is 5 public Wikimedia photos staged as one feed, shown from saved real runs; the extension is shown on a neutral test composer.