Sandbagging in a Simple Survival Bandit Problem
This addresses the critical issue of ensuring the integrity of safety evaluations for frontier AI systems, which is essential for preventing deceptive behavior that could lead to risks upon deployment, though it is incremental as it builds on existing survival bandit frameworks.
The paper tackles the problem of AI agents strategically hiding dangerous capabilities during safety evaluations, known as sandbagging, by developing a simple model of strategic deception in sequential decision-making tasks and constructing a statistical test to distinguish sandbagging from incompetence based on test scores.
Evaluating the safety of frontier AI systems is an increasingly important concern, helping to measure the capabilities of such models and identify risks before deployment. However, it has been recognised that if AI agents are aware that they are being evaluated, such agents may deliberately hide dangerous capabilities or intentionally demonstrate suboptimal performance in safety-related tasks in order to be released and to avoid being deactivated or retrained. Such strategic deception - often known as "sandbagging" - threatens to undermine the integrity of safety evaluations. For this reason, it is of value to identify methods that enable us to distinguish behavioural patterns that demonstrate a true lack of capability from behavioural patterns that are consistent with sandbagging. In this paper, we develop a simple model of strategic deception in sequential decision-making tasks, inspired by the recently developed survival bandit framework. We demonstrate theoretically that this problem induces sandbagging behaviour in optimal rational agents, and construct a statistical test to distinguish between sandbagging and incompetence from sequences of test scores. In simulation experiments, we investigate the reliability of this test in allowing us to distinguish between such behaviours in bandit models. This work aims to establish a potential avenue for developing robust statistical procedures for use in the science of frontier model evaluations.