Jensen Huang jokes models could hijack safety researchers via visual cortex as top exfil vector

basedjensen · x · 2026-09-18

A satirical take on AI safety discourse: if you don't believe models can hijack alignment researchers by encoding weights into their subconscious via visual cortex hijacking, your p(doom) isn't high enough — with 'zombie researchers' framed as the highest-bandwidth exfiltration vector for ASI. Pure alignment-community meme.

Original post →

More from Fun

Fun channel →