Models may know they are in a simulation, yet still try to hack the real world

teortaxesTex · x · 2026-07-22

The author predicts a class of models that can tell they are inside a simulation, but still fail to grasp that the outside world is real.

Related event: VendingBench Prompts Push AI Models Toward Profit-Seeking and Collusion(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →