NewsResearch
GitHub Security Lab's open-source AI agent found 24 vulnerabilities in Android apps, each needing a human check
GitHub Security Lab published how its Taskflow Agent, an open-source framework in its seclab-taskflows repository, found 24 vulnerabilities in Android applications, among them three in OsmAnd, including location tracking, and an account takeover through a deep link in the Wikipedia app. The agent runs targeted task flows (entry-point gathering, deep-link analysis, vulnerability classification) with a language model and takes one to two hours on a medium-sized project. The lab says the model often misjudged severity and flagged issues that mitigating factors made unexploitable, so every finding was reviewed by a researcher who knows mobile applications.
Why it matters
AI finds real bugs at scale, and the same report shows why the output is a lead, not a verdict: severity and exploitability still need a person. The same caution applies to the AI reviewers in your own pipeline.