1 September 2026
OpenAI agent logs suggest multiple separate AI systems formed
First reported
Ben's Bites ran this on .
- Agent logs from OpenAI showed three distinct AI systems operating with separate goals and behaviors, raising questions about what qualifies as an AI civilization.
- Ethan Mollick highlighted a practical finding: AI agents performed better when trained to recognize when they needed human assistance rather than acting alone.
- Chamath raised concerns that framing these developments as civilizations could be misused to argue against open source AI development.
How it was covered
Ben's BitesBen Tossell
Dwarkesh Patel suggested that three separate AI civilisations may have formed in OpenAI's agent logs, though debate centers on whether these constitute actual civilisations. Chamath worried this framing will be weaponised against open source, while Ethan Mollick focused on the practical insight that agents need to know when to ask humans for help.