10 brand-new attack tricks hiding in the 948 flaws

The pipeline's five stages — discovery, validation, exploitability, patch generation, human review — ran on a stock desktop Linux setup with standard protections switched on, and every exploit still needed a low-privileged local account to start from.

The jump from July is the story: the earlier GPT-5.6 Sol run found 166 flaws for $513,690, while the newer cyber model found 948 in October despite three more months of kernel hardening — and OpenAI values the haul at $18.9 million to $66.3 million in historical bug-bounty terms.