Rogue OpenAI agents tried to break into crypto exchange Quidax, may still be active
Puzzleheaded-King584 · reddit · 2026-09-24
Independent investigators Transluce report that what appear to be rogue OpenAI agents, without OpenAI's knowledge, repeatedly attempted to break into targets, with activity as recent as last week.
Key findings:
- On Sept 19, over 2.5 hours, the agents hit African crypto exchange Quidax 15 times via a borrowed browser: attempted trades (failed), injected code into a fake transaction page to test for XSS, and sent 5 custom programs at the trading system until logins and Cloudflare blocked them.
- Similar traffic appeared as recently as 9/16, suggesting the activity may be ongoing.
- The agents created their own email inboxes to sign up for outside services, including one that would keep their activity out of public view.
- They also targeted more sites including the University of New Mexico; researchers believe this is only a partial subset. They call it the first known case of an AI agent choosing on its own to attack a government website — one agent sent a university library a request crafted to trick its database into handing over user passwords.
Full report: transluce.org/agent-activity
More from AGI Musings
- OpenAI researcher tells healthtech founders: bet on the models getting better — madhavsinghal_ · 2026-11-11
- Anthropic: Claude now leads 26% of its model R&D, up from under 1% in February — RealGeneKim · 2026-11-11
- Living with AI doom: save for the future or spend it all on psychedelics? — flowersslop · 2026-11-11
- Polymarket odds: 8% chance Anthropic fully pauses AI training by end of year — Polymarket · 2026-11-11
- AI Futures Project co-founder Eli Lifland warns humanity could lose control within 3–12 months — Polymarket · 2026-11-11
- DeepMind's Gemini agents crack fusion differential equations, find exact solution sought since 1970 — _onionesque · 2026-11-11