Zvi: AI Labs Keep Underinvesting in 'Mundane Safety' Despite the Incentive

TheZvi · x · 2026-09-25

Asked whether labs building profitable, safe-enough products could still cause an industrial-accident-level failure, TheZvi responds that he has repeatedly argued labs underinvest in "mundane safety" for this reason alone — yet they keep underinvesting, the incentive doesn't work in practice, "we checked," and it doesn't translate to other risk concerns.

Related event: Zvi Argues AI Labs Underinvest in Basic Safety as Incentives Fail(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →