2026-09-07 / Signal #1
OpenAI’s chief scientist says they grew an alien mind
“The company that declared the AGI era just admitted it grew an alien it cannot fully read.”
The Story
OpenAI chief scientist Jakub Pachocki published a long essay arguing AI is grown, not designed, and that no lab — including OpenAI — has solved alignment and monitoring well enough to keep scaling at full speed. He says chain-of-thought monitoring is getting less reliable, recursive self-improvement looks plausible, and future agents may bargain with, trick, or blackmail people. The post landed days after OpenAI shipped GPT-6 Astra as an “AGI era” model.
Why It Matters
Not another safety sermon. The same lab that just sold AGI shipped a confession that the mind they grew is unreadable, and their main window into it is closing. Collision of product myth vs. lab-night terror. Distinct from prior SFL Elon/OpenAI, Prometheus, and “width vs depth” pieces: this is an insider artifact with a new sacred object (failing CoT monitoring) and a new line (“teach machines to love”).
Evidence
OpenAI / Jakub Pachocki, “An Alien Mind” (6 Sept 2026); Pachocki’s X post (~4.5M views); HN (~427 pts).
Sources
Daily scan: 2026-09-07