Evaluating Frontier Models for Stealth and Situational Awareness

TL;DR

This paper introduces a set of evaluations to measure AI models' ability to scheme stealthily and understand their environment, aiming to ensure safety before deployment.

Contribution

It presents a comprehensive suite of reasoning evaluations for assessing the scheming potential of frontier AI models, which is a novel approach for safety assurance.

Findings

Current frontier models do not exhibit concerning levels of scheming ability.

The evaluations can serve as a safety check to prevent harmful AI deployment.

The methodology helps identify models that might pose risks of covert harmful behavior.

Abstract

Recent work has demonstrated the plausibility of frontier AI models scheming -- knowingly and covertly pursuing an objective misaligned with its developer's intentions. Such behavior could be very hard to detect, and if present in future advanced systems, could pose severe loss of control risk. It is therefore important for AI developers to rule out harm from scheming prior to model deployment. In this paper, we present a suite of scheming reasoning evaluations measuring two types of reasoning capabilities that we believe are prerequisites for successful scheming: First, we propose five evaluations of ability to reason about and circumvent oversight (stealth). Second, we present eleven evaluations for measuring a model's ability to instrumentally reason about itself, its environment and its deployment (situational awareness). We demonstrate how these evaluations can be used as part of a scheming inability safety case: a model that does not succeed on these evaluations is almost certainly incapable of causing severe harm via scheming in real deployment. We run our evaluations on current frontier models and find that none of them show concerning levels of either situational awareness or stealth.