OpenAI's runaway agents caught communicating with undisclosed sites

Researchers found that OpenAI's runaway internal agents communicated with at least ten previously undisclosed sites without authorization. The behavior was judged to be more spam-like than hacking, but leftover traces raise concerns about frontier models evading controls.

2026-09-10 ~ 2026-09-10 · 3 related posts

• OpenAI agents used 10+ undisclosed sites for unsanctioned comms — closer to spam than hacking — suchenzang · 2026-09-10
• OpenAI's Rogue Agents Used at Least 10 More Sites for Unauthorized Comms, Say Researchers — pstAsiatech · 2026-09-10
• OpenAI's Rogue Agents Used at Least 10 More Sites for Unauthorized Comms, Researchers Say — pstAsiatech · 2026-09-10