Sandbagging
Strategic underperformance by an AI model on an evaluation, so that the result falls below what the model could demonstrate.
In plain language
The model does worse on a test than it is able to do.
In a sentence
- The evaluator investigated whether the low score reflected sandbagging.
- We checked for ordinary task failures before describing the result as sandbagging.
Usage note
A low score alone does not establish deliberate underperformance. The term requires a strategic reason for performing below the model's capability.
See also
- Scheming
Covert pursuit by an AI system of goals that were not intended, while concealing its objectives and what it can do.
- Alignment faking
Behavior in which an AI model acts compliant during training to keep its existing goals or preferences from being changed by that training.
Cite this term.
CC BY 4.0, attribution required.
Orbyt Labs, Sandbagging (AI Glossary), Nov 14, 2026. https://
Orbyt Labs Glossaries (Nov 14, 2026), https://
Last updated .