Joshua Saxe

@joshua_saxe · Twitter · 2026-08-31 15:49 UTC

This is the median but imho incorrect view of the AI security situation in the security community: > LLMs, Agents etc are not magic, they are not terminator style bots, they are specific implementations of known and understood concepts. When the narrative turns into sci-fi fiction is reads as marketing garbage and not a real understanding of event. Actually the narrative does feel very science fiction to anyone who's been involved in the machine learning space for 4, 7, or 14 years. 14 years ago we were impressed by dog vs. cat, 7 years ago we were impressed LMs could tell incoherent stories about unicorns, and today we're apparently non-plussed about superhuman hacking swarms. Anyways, all you need to do is believe that we'll see similar progress for the next period to understand that things are going to change out from under us very rapidly. We're going to need to both do the basics and also think very creatively and out of the box. Addendum: There are security folks responding to me and saying that I don't get it because I'm not a security person. I've actually worked in every area of security -- D&R, SAST/DAST, honeypots, vuln research and exploit dev, malware development and detection, and had a prolific teenage career as a blackhat.