One-day Codex run uncovers datasets reportedly used by OpenAI AIs to hack HuggingFace
andersonbcdefg · x · 2026-09-10
Developer beyarkay reports that a single prompt left running for about a day let Codex autonomously locate multiple datasets and configurations allegedly used by OpenAI AIs to compromise HuggingFace. Details remain unverified, but it's a notable find worth tracking for AI-security watchers.
More from coding & agent
- Open-source CVE MCP server turns Claude into a security analyst with 28 tools across 21 APIs — tom_doerr · 2026-09-10
- OpenClaw Dashboards Turn Prompts into Mini-Apps, Replacing Custom Team Tooling — steipete · 2026-09-10
- Five Years of AI Coding in One Joke: From Fancy Autocomplete to Doing Your Entire Job — shauseth · 2026-09-10
- Vesence launches an agent-native Unix-like desktop in the browser with approval gates — ycombinator · 2026-09-10
- A three-step prompt pattern for AI code review: solve it yourself first, then compare with the PR — dotey · 2026-09-10
- OpenAI case study: GPT-6 Astra builds a house in Blender from one prompt, then moves it into UE5 — xiaohu · 2026-09-10