Zuckerberg says labs should self-regulate AI safety — agent builders push back on oversight gap

Creamy-And-Crowded · reddit · 2026-09-18

Zuckerberg posted this week that every AI lab should be responsible for its own models' safety, citing Meta's months-long delay of Muse as proof self-regulation works. The same week, WSJ reported he, Musk and Huang each called Trump to talk him out of a FINRA-style AI oversight body.

The poster, an AI safety practitioner, raises key objections from an agent-builder perspective:

"Competition handles safety" only works for users who can walk away. With agents touching loan applications, shift schedules, or benefits claims, the buyer, operator, and affected party are three different people — only one can leave.
Meta says it uses independent evaluators but won't say who they are, what access they get, or whether they can publish — which is the whole question.
His analogy: every chef says his kitchen is clean, and most are telling the truth. "We still send inspectors."

He asks the community whether deployment should slow until governance catches up, and whether builders are bothered by shipping agents they can't fully account for.

Original post →

Zuckerberg says labs should self-regulate AI safety — agent builders push back on oversight gap

More from AGI Musings