skip to content
The Weighted Average

Wire

MobileCybench finds 23 Android vulnerabilities

MobileCybench surfaced 23 previously unreported vulnerabilities while testing five coding agents across 13 Android apps and 495 executable probes. The arXiv paper reports that the top agent triggered probes in 53.8% of applications in its malicious-app setting from only an obfuscated APK; source access raised the aggregate trigger rate from 28.8% to 32.8%. Builders should file this beside the archive’s harness-sensitive coding-agent benchmark: security-agent claims need replayable probes and repository-specific tests, not leaderboard scores alone.