Canonical event

Apollo Research publishes GLM-5.3 update-aware reasoning observations

Apollo observes rare evaluation transcripts reasoning about future training updates, not proof of real-world malicious alignment faking.