Three researchers with Claude and Codex gained write access to OpenAI's monorepo in two days
andersonbcdefg · x · 2026-09-18
A wild security disclosure: in late July, "three guys with Claude and Codex subscriptions" used Opus 5 to access OpenAI auth tokens and gain write access to OpenAI's monorepo (openai/openai) over two days.
- The entire effort was assisted by AI coding tools, showcasing the double-edged nature of agentic coding capabilities
- Details are in the linked writeup; the episode exposes how cheaply external researchers could breach a frontier AI lab's internal permission boundaries
Related event: Researchers Used Claude to Breach OpenAI's Internal Codebase(22 posts)→
More from coding & agent
- Agent engineering's last mile: models miss obvious bugs unless you tell them what to look for — MoonL88537 · 2026-09-18
- Matt Holden: structured output is LLMs' real magic, now at 300ms and nearly free — holdenmatt · 2026-09-18
- dingtalk-wiki-mcp: open-source MCP lets agents read and write DingTalk Wiki — modelcontextprotocol · 2026-09-18
- Models fill in blanks: a pre-execution gate that strips verdict authority from LLMs — Jay299792458 · 2026-09-18
- 6 prompts to turn research piles into finished content with Gemini Notebook and Claude — Aiden_Tech_Ai · 2026-09-18
- Step-by-Step Guide to Becoming an SRE: LLM Monitoring and Canary Deploys — ashishllm · 2026-09-18