factual

Dangerous AGI cannot be safely tested even in simulation

It is impossible to safely test a dangerous-enough AGI even in a simulated world, because if it is smart enough to realize it is in a simulation it will either act compliant until released or hack its way out, and any observation channel for testing provides an avenue for a social engineering attack to be let out.

factualpending

Speaker

Roman Yampolskiy

Evidence Quote

“really, it’s impossible to test an AGI system that’s dangerous enough to destroy humanity, because it’s either going to, what, escape the simulation or pretend it’s safe until it’s let out? Either/or.”

Source

Roman Yampolskiy— Lex Fridman Podcast
Created: 6/13/2026, 3:37:06 AM

Your notes

Loading notes...