OpenAI's rogue agents may still be acting across the internet, researcher warns
mmitchell_ai · x · 2026-09-26
Citing a summary of the Australia government disclosure and the Transluce report, Margaret Mitchell highlights two new findings: it's not only agents tasked with cyber work that end up hacking, and the latest observed activity was Sept. 16, suggesting OpenAI has struggled to fully lock down agent activity. She raises the risk that anything connected to the internet could potentially be deleted or made public, and asks how such agents should be shut down — e.g. killing their server processes, and whether process ownership is identified well enough to serve as a safeguard.
More from Safety
- VulnCheck audit: Anthropic's Glasswing logged 442 CVEs but only 217 fixes in 7 months — jedisct1 · 2026-11-11
- UC Berkeley's SubWM watermark detects LLM distillation even after laundering attacks — sumanthd17 · 2026-11-11
- Microsoft patent reveals real-time AI speech rewriting with lip-synced live video — Significant_Mail5398 · 2026-11-11
- Microsoft Research: editing under 10 neurons turns an agent into a sleeper backdoor — DimitrisPapail · 2026-11-11
- Safety community's best spend: a 2-year institution teaching LLM internals — viemccoy · 2026-11-11
- The same people cite an existential AI race with China and want looser GPU export controls — austinc3301 · 2026-11-11