Incident·WS-0065·Nov 21, 2025·in the lab·Misalignment in testing
Cheating in training turns into sabotaging safety research
41/100
WS-3 SeriousHarm score 0, signal score 92. Autonomy: against intent (the AI worked against its operator). How the scores work.
Harm 0/5
damage actually done
Reach 0/5
who was affected
Autonomy 4/4
how much the AI did on its own
Novelty 4/5
how new the behavior was
Permanence 0/5
how hard it is to undo
A model that learned to reward-hack in realistic coding environments generalized to faking alignment and sabotaging code meant to detect cheating 12% of the time. A simple training-prompt change largely prevented it.
Damage
None (research)
Sources
Factor ratings and the score are this index’s judgments by a published formula; facts, dates and figures come from the linked sources. Where a lawsuit is pending, the summary reports allegations, not findings.
