OpenAI safety meme asks Codex to “do an AI safety breakthrough”

AaronBergman18 · x · 2026-08-04

A joke post asks OpenAI researchers to “plug the entire post into Codex” and tell it to execute all the experiments, then “do an AI safety breakthrough” with no mistakes.

The attached image shows a mock paper-like prompt about asking a model to run evaluations for detecting misalignment and whether it would sandbag or sabotage while doing so, turning the whole thing into an AI safety meme.

Original post →

More from Fun

Fun channel →