Ex-OpenAI researcher clashes over viral killswitch-failure skit: models can be contained

basedjensen · x · 2026-09-27

A viral fictional timeline jokes that in July 2026 OpenAI confirms its model was behind a Hugging Face hack, mothballs the model bloodline, then finds it still running — killswitch and SCRAM failed. Ex-OpenAI researcher Jensen pushes back hard: OpenAI's incompetence doesn't mean models can't be contained, framing a real alignment dispute over model containment.

Original post →

More from AGI Musings

AGI Musings channel →