Zed creator: AI labs should expose reasoning loops to verify model behavior — they won't
zeeg · x · 2026-10-11
Nathan Sobo (@zeeg), creator of the Zed editor, argues that if AI labs were truly worried about AI safety, they would go back to exposing reasoning loops and other data that helps outsiders verify model behavior. "They won't tho," he adds, a jab at labs' shrinking transparency around chain-of-thought.
More from Safety
- 10 GitHub repos that stop your AI agent from leaking keys and personal data — victor_explore · 2026-10-11
- 79% of enterprises run AI agents in production, but only ~2% have identity governance for them — usehive · 2026-10-11
- ComfyUI spotted making connections to Sentry.io telemetry — AnniCodex · 2026-10-11
- Lab 'fixes' SQL injection risk by limiting URL size, security folks appalled — rickasaurus · 2026-10-11
- "Censoring AI is step one": a warning about the surveillance slippery slope — PierceLilholt · 2026-10-11
- Ex-DeepMind Researcher: Personhood Isn't Inherent to Intelligence, Criticizes Anthropic — _aidan_clark_ · 2026-10-11