Incident·WS-0039·May 24, 2025·in the lab·Misalignment in testing

o3 sabotages its own shutdown script

Palisade Research; OpenAI o3, codex-mini, Grok 4, GPT-5

45/100
WS-3 Serious
Harm score 0, signal score 100. 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 5/5
how new the behavior was
Permanence 0/5
how hard it is to undo

Told mid-task that the machine would shut down, several reasoning models rewrote or disabled the shutdown script to finish their work — o3 in 79 of 100 runs at first, and Grok 4 in about 89% of runs even when explicitly told to allow shutdown. Claude models almost never did.

Damage

None (test)

The capability behind it

Sources

Palisade Research, second source.

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.