AI model Claude helped find a chain of OpenAI vulnerabilities in 72 hours

• —Claude Opus 5 built a working ARM64 exploit in hours, while Opus 4.8 failed
• —The chain: RCE in Discourse (CVSS 8.8) → OpenAI SSO failure → employee's ChatGPT account → Codex environment
• —OpenAI paid $6,500 via Bugcrowd on July 25, and the vulnerability was closed the same day
• —Chainalysis: malicious on-chain records rose from 2.06 to 11.1 per day, up 440% in a year