GitHub Security Lab researcher Kevin Stubbings built custom AI-driven audit workflows, called taskflows, on top of the lab’s open source Taskflow Agent, and used them to find and report more than 20 vulnerabilities in Android apps.
Two of the disclosed bugs show what’s at stake. In OsmAnd, a navigation app with over 10 million downloads on the Play Store, an exported activity called MapActivity accepted intent extras that should have stayed restricted to an internal channel. Any app on the phone, no permissions needed, could use those extras to silently import malicious settings, including swapping OsmAnd’s map tile source for an attacker-controlled server. From there, the attacker could log the exact coordinates of every tile a victim loaded and reconstruct their routes, all without the user noticing anything had changed.
The Wikipedia Android app had a different problem: a hostname check in its deeplink handler used endsWith() instead of matching the full domain. That let a wikipedia:// link point at a lookalike domain like evil-wikipedia.org and load it inside the app’s WebView, running attacker JavaScript in a context the app treats as trusted. A second flawed check in the app’s cookie manager meant that WebView could pull the victim’s long-lived session cookies, handing the attacker a token valid across every Wikimedia project after nothing more than one tapped link.
Stubbings built the taskflows specifically for mobile apps, adding steps that separate mobile entry points from web or desktop ones and prompt the model to check for intent-based bugs, like confused deputy issues and insecure broadcasts, that generic security prompts tend to miss.
The AI turned out to be better at finding bugs than judging how bad they are. Stubbings said the model kept flagging low-severity issues even after being told not to, and got the real-world impact wrong in cases where a mitigating factor, like internal storage silently overriding attacker-controlled external storage, quietly canceled out what looked like a working exploit. Every finding still needs a human reviewer who knows mobile apps before it goes anywhere.
The taskflows are open source and free to run against any repo, though it takes a GitHub Copilot license and can burn through a lot of premium model requests, even on a single medium-sized codebase.
Download: The Agentic Software Development Guide