New Study Shows GPT-5.2 Can Reliably Develop Zero-Day Exploits at Scale

**Executive summary:** A security researcher tested GPT-5.2 and Opus 4.5 on a QuickJS zero-day and found the models produced over 40 working exploits across multiple configurations, including sophisticated chains that bypassed modern mitigations; the study warns that automated exploit generation is becoming computationally feasible and could shift offensive capability metrics from human expertise to compute and token budgets.