AI security taskflows reportedly helped find 24 Android app vulnerabilities

The GitHub Blog says its team found and reported 24 Android app vulnerabilities using targeted taskflows for an open-source AI security agent. It describes an OsmAnd flaw that could let another app silently change settings to expose a user's location, and two Wikipedia app bugs that could be chained to steal cookies and take over an account. The Blog cautions that researchers need to review AI findings because the agent can misjudge severity and return false positives.