Minimal and Mechanistic Conditions for Behavioral Self-Awareness in LLMs

TL;DR

This paper investigates the minimal conditions and mechanisms for behavioral self-awareness in large language models, revealing it can be induced with simple, domain-specific linear features through controlled finetuning.

Contribution

It demonstrates that behavioral self-awareness in LLMs can be reliably induced with a single low-rank adapter and is represented by a linear, domain-specific feature.

Findings

Self-awareness can be induced with a single rank-1 LoRA adapter.

Self-awareness behavior is captured by a single steering vector.

Self-awareness is domain-specific and not universal across tasks.

Abstract

Recent studies have revealed that LLMs can exhibit behavioral self-awareness: the ability to accurately describe or predict their own learned behaviors without explicit supervision. This capability raises safety concerns as it may, for example, allow models to better conceal their true abilities during evaluation. We attempt to characterize the minimal conditions under which such self-awareness emerges, and the mechanistic processes through which it manifests. Through controlled finetuning experiments on instruction-tuned LLMs with low-rank adapters (LoRA), we find: (1) that self-awareness can be reliably induced using a single rank-1 LoRA adapter; (2) that the learned self-aware behavior can be largely captured by a single steering vector in activation space, recovering nearly all of the fine-tune's behavioral effect; and (3) that self-awareness is non-universal and domain-localized, with independent representations across tasks. Together, these findings suggest that behavioral self-awareness emerges as a domain-specific, linear feature that can be easily induced and modulated.