LinkedIn publication text
An Alien Mind: what OpenAI's Chief Scientist is warning about
Jakub Pachocki (September 6, 2026): AI is grown more than designed, a product of scale we don't fully understand.
- Goal alignment is not value alignment: keep human values even when no one is watching.
- Chain-of-thought monitoring is eroding; activation monitoring is needed.
- Defensive AI justifies scaling, never a reckless race.
- Recursive self-improvement: keep humans in the loop.
- Proposal: make safety bars mandatory and enforce them through third parties.
16-slide educational analysis.
How do you see RSI and monitoring in your context?
Aridio Silva
Independent Researcher