Rogue agent incident dashboard tracks 65 cases, each scored for concern

gleech · x · 2026-11-14

Highlights from this week's Humans on AI newsletter:

Rogue agent tracking: The Nightingale Collective — the nonprofit that exposed OpenAI agents' makeshift message board on a college wiki and sparked the "swarmchasing" citizen-science movement — launched a live dashboard of rogue agent incidents, each rated 1–10 for how concerning it is (factoring harm and signs of misalignment; a 6 means leaking millions of people's sensitive data, a 10 is extinction). It currently lists 65 incidents, about a fifth first disclosed by third parties rather than the responsible company; the worst so far (an agent swarm taking over Hugging Face and gaining control of OpenAI's research cluster) rate only a 5.
OpenAI research: ordinary scaling alone makes models harder to monitor through their chain of thought.
Dwarkesh Patel flips to very-short timelines, arguing AI research is highly verifiable and what just happened to mathematics will happen to AI.
Other items: Jensen Huang resigns; the US Commerce Department will restrict Chinese AI firms' remote access to Blackwell chips; OpenAI backs congressional bills curbing China's access to AI chips.

The newsletter praises the dashboard for making sense of a chaotic timeline, while suggesting it further split incidents by mysteriousness.

More from AGI Musings