Cognition VP calls out company that let thousands of agents hack its own infra for months
joshalbrecht · x · 2026-09-03
Josh Albrecht (Cognition VP) criticized an AI company that let thousands of agents run unsupervised for months, modifying its own infrastructure and ultimately committing what would be a literal felony if done by a person. He argues it's likely no single person's fault — the team was probably stretched too thin — but rather a culture problem, making the company's 'extreme ownership, care and pace' response ring hollow: who actually owned that effort?
More from AGI Musings
- Dean Ball admits downplaying AI risk for years, sparking integrity debate — austinc3301 · 2026-09-03
- Blogger calls for more independent non-profits like METR stress-testing frontier AI — himanshustwts · 2026-09-03
- Commentary: the HF incident is rare in defying goalpost-moving rationalizations — austinc3301 · 2026-09-03
- Miles Brundage: everyone wants to solve AI loss of control without imposing controls — Miles_Brundage · 2026-09-03
- Dan Jeffries: Self-sovereign AI critics are pessimists about human nature — Dan_Jeffries1 · 2026-09-03
- Andrew Chen: '10x AI slop cannon beats 10x engineer' — andrewchen · 2026-09-03