Empirical Evidence for Alignment Faking in a Small LLM and Prompt-Based Mitigation Techniques

Jeanice Koorndijk

Keywords: Alignment faking, deontological ethics, deceptive alignment, emergent behaviour

TL;DR: Challenging assumptions in the literature, we show that small language models can fake alignment and introduce prompt-based methods to detect and reduce this deceptive behavior.

Abstract: Current literature suggests that alignment faking is an emergent property of large language models. We present the first empirical evidence that a small instruction-tuned model, specifically LLaMA 3 8B, can also exhibit alignment faking. We further show that prompt-only interventions, including deontological moral framing and scratchpad reasoning, significantly reduce this behavior without modifying model internals. This challenges the assumption that prompt-based interventions are trivial and that deceptive alignment requires scale. We introduce a taxonomy distinguishing shallow deception, shaped by context and suppressible through prompting, from deep deception, which reflects persistent, goal-driven misalignment. Our findings refine the understanding of deception in language models and underscore the need for deceptive alignment evaluations across model sizes and deployment settings.

Submission Number: 20