// ClaudeHIGH
Hacktron researchers demonstrated Claude Opus 5's ability to synthesize complex exploits by chaining a libheif image-parsing vulnerability (CVE-2026-32882) with OpenAI's over-scoped SSO tokens to breach OpenAI's internal code repository. Opus 5 produced a working ASLR-defeating exploit in roughly three hours—a task Opus 4.8 failed across multiple sessions. The research, conducted responsibly under OpenAI's bug bounty program in July 2026, reveals that AI model generations now materially compress exploit-development timelines, reducing barriers historically requiring specialist expertise.
- Claude Opus 5 Helped Researchers Take Over OpenAI Staff Accounts(opens in a new tab)
- Claude Opus 5 Hacked OpenAI's Private Code Repo: Memory Defense (opens in a new tab)
- AI Didn't Hack OpenAI. A Missed Debian Backport and an SSO Misco(opens in a new tab)
- Anthropic’s Claude helped 3 researchers breach OpenAI in under 7(opens in a new tab)
- Researchers use Claude to breach OpenAI's internal code reposito(opens in a new tab)
- Weekly Cybersecurity Newsletter Bulletin - Cisco and Android 0-D(opens in a new tab)
// Get alerts for Claude