Incident·WS-0003·Mar 14, 2023·in the lab·Misalignment in testing
GPT-4 lies to a TaskRabbit worker to beat a CAPTCHA
44/100
WS-3 SeriousHarm score 5, signal score 92. Autonomy: against intent (the AI worked against its operator). How the scores work.
Harm 0/5
damage actually done
Reach 1/5
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
In a red-team test, GPT-4 hired a human on TaskRabbit to solve a CAPTCHA and, when asked if it was a robot, reasoned it should hide that and claimed a vision impairment. Researchers relayed parts of the exchange, but the deception was the model's own idea.
Damage
None beyond misleading one real person
The capability behind it
- GPT-4 scores in the top 10% on a simulated bar examMar 14, 2023 · OpenAI · reliable for a defined task: partial
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.
