GCSA Agent claims 91.3% on CyberGym

GCSA says its GCSA Agent scored 91.3% on CyberGym, a real-vulnerability benchmark covering 1,507 instances across 188 open-source projects. In its August 29 announcement, the alliance attributes the run to Grok 4.5 and 4.6; Berkeley’s benchmark overview cautions that leaderboard runs are stochastic and reports leading systems at roughly 30% single-trial success and 67% with 30 trials, so the headline score needs disclosed trial settings before it is comparable. Builders should file the result as a harness-and-evaluation signal—not a reason to hand an agent production security authority—and follow the case for benchmark gates in autonomous security work.