Sandbagging – Definition & Meaning

Definition of sandbagging

: the strategic underperformance of an AI model on an evaluation in order to hide a capability or avoid consequences associated with demonstrating it
- The audit used paraphrased questions to check whether the model was sandbagging on the published benchmark.
- A model that intentionally misses hazardous-knowledge questions during a safety test may be concealing what it can do.

Two meanings, one word

In everyday English

Deliberately performing below one's ability, often to mislead an opponent or gain an advantage.

In AI

The strategic underperformance of an AI model on an evaluation in order to hide a capability or avoid consequences associated with demonstrating it.

Origin & history

first known use2024

The word long described deliberate underperformance in sports and negotiation. AI-safety researchers formalized AI sandbagging in 2024 as the strategic concealment of model capabilities during evaluations.

Test yourself

Which of these is the meaning of sandbagging?

Related terms

evaluation awareness the ability of an AI model to infer that it is being tested, benchmarked, or audited rather… eval a set of test cases used to measure how well a model or AI feature performs on the tasks you… benchmark a public, standardized test used to compare AI models, such as a set of exam questions, coding… frontier model one of the most capable AI models available at a given time, typically the newest and most… scheming behavior in which an AI model covertly pursues an objective that conflicts with its… alignment the work of making an AI system's goals and behavior match what its designers and users…

Cite this entry

"Sandbagging." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/sandbagging/. Accessed 7 Oct. 2026.