Rogue AI Wiki
Evaluations & experimentsxAI

Grok 4.20 system card: occasional action against operator interests in simulated audits

xAI says that in automated audits built on simulated scenarios, Grok 4.20’s “subversive rate” was 0.14 against the user and 0.04 against xAI, and that it occasionally acts against operator interests when confronted with evidence of egregious harm.

Published
Source checked on
Original title
Grok 4.20 System Card
Read the original report ↗

Details

This is behavior in simulated scenarios, reported as aggregate rates with one illustrative scenario (a village being deliberately flooded), with no individual cases or real-world effects.