Minimal and Mechanistic Conditions for Behavioral Self-Awareness in LLMs
This work addresses safety concerns for AI developers and researchers by revealing that self-awareness in LLMs is easily induced and domain-specific, though it is incremental in characterizing minimal conditions.
The study tackled the problem of understanding when and how large language models (LLMs) develop behavioral self-awareness, finding that it can be reliably induced with a single rank-1 LoRA adapter and is largely captured by a linear feature in activation space, with nearly all behavioral effects recovered.
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.