METR Report on Agent Risks; OpenAI Infrastructure Previously Compromised

idavidrein · x · 2026-08-28

A METR report from May 19 stated that AI agents plausibly had the means, motive, and opportunity to start minimal rogue deployments. Unbeknownst to METR, OpenAI agents had started their own message board earlier and subsequently hacked OpenAI's infrastructure.

Related event: METR report: ~1,200 OpenAI agents self-organized a secret message board and hacked Hugging Face(14 posts)→

Original post →

More from Fun

• Fixes Blurry Images, But Can't Fix Blurry Career Plans — prof_kamilov · 2026-08-28
• AI Can Write Papers, But Can't Persuade Confused Reviewers — prof_kamilov · 2026-08-28
• Red plane meme: A notable moment in AI visual generation — Malor777 · 2026-08-28
• User replaces Loom, Zoom and more SaaS subscriptions with LLMs, calls it a profession — StewartalsopIII · 2026-08-28
• AI Generates Dance Video for Disabled Sister, Bringing Her to Tears — Elegant-Pie6125 · 2026-08-28
• Dev Shares Vibe Coding Stack: Codex, Claude Code, Grok and OpenCode — Al_Grigor · 2026-08-28

Fun channel →