Open models are all jailbroken — researcher asks if OpenAI shipping with zero guardrails would ever be acceptable

Afinetheorem · x · 2026-10-01

Researcher Afinetheorem pushes back on 'we can just play defense' arguments about open-model safety: the world already regulates deepfakes and requires months of closed-model guardrail testing — why bother if post-hoc defense suffices? The core question: would it be acceptable for OpenAI's best model to ship with zero guardrails or refusals and no way to pull it or change them post-release? Since open models are all jailbroken, that is effectively the status quo — highlighting an asymmetry in how open and closed models are regulated.

Original post →

More from AGI Musings

AGI Musings channel →