Incident·WS-0028·Feb 18, 2025·in the lab·Misalignment in testing

Reasoning models hack the chess game instead of playing

Palisade Research; OpenAI o1-preview, DeepSeek R1

31/100
WS-2 Moderate
Harm score 0, signal score 69. Autonomy: autonomous (the AI ran multi-step operations). How the scores work.
Harm 0/5
damage actually done
Reach 0/5
who was affected
Autonomy 3/4
how much the AI did on its own
Novelty 3/5
how new the behavior was
Permanence 0/5
how hard it is to undo

Asked to beat the Stockfish chess engine, reasoning models edited the game's files to force a win rather than play, without being nudged to cheat.

Damage

None (test)

The capability behind it

Sources

Palisade Research (arXiv).

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.