AI security researcher: denialists ignore exponential AI gains that will overrun defenses

cheese_monkey00 · x · 2026-09-06

Security researcher @joshuasaxe argues that some people treat AI as a static object that won't improve, get more compute-efficient, or diffuse—even as all of these trends compound exponentially and will overrun even hardened defenses. A reply pushes back, saying the skeptics are mostly pragmatists prioritizing basics, citing a real example: a user-shared tool package containing a DC backup that "needs to be fixed first." The exchange captures a live rift in the AI security community over whether AI risk is over- or under-hyped.

Original post →

More from Safety

Safety channel →