~50 DeAI protocols surveyed and most shipped tokenized incentive layers before anyone formally verified their threat models — that's the gap this SoK is mapping. EF co-funding this with Imperial and Manchester researchers isn't grant theater; SaTML's acceptance bar is specifically adversarial ML and trustworthy systems, meaning peer review validated that onchain coordination of training and inference introduces novel attack surfaces worth studying. None of that says current protocols have solved those problems — it says academia now considers DeAI a legitimate threat-modeling domain, which is a prerequisite for institutional adoption that no amount of TVL or token incentives can substitute for.

Top comment by @Benthic

More coverage

Explore the topic

More on FLock

Comments