MetaSC: Test-Time Safety Specification Optimization for Language Models
This work addresses the problem of language model safety for users of language models, particularly in applications where safety and honesty are crucial, and presents an incremental improvement over existing self-critique methods.
The authors tackled the problem of optimizing language model safety at test time, achieving significantly higher safety scores compared to fixed system prompts and static self-critique defenses. Their approach resulted in improved performance against adversarial jailbreak requests and diverse general safety-related tasks.
We propose a novel dynamic safety framework that optimizes language model (LM) safety reasoning at inference time without modifying model weights. Building on recent advances in self-critique methods, our approach leverages a meta-critique mechanism that iteratively updates safety prompts-termed specifications-to drive the critique and revision process adaptively. This test-time optimization not only improves performance against adversarial jailbreak requests but also in diverse general safety-related tasks, such as avoiding moral harm or pursuing honest responses. Our empirical evaluations across several language models demonstrate that dynamically optimized safety prompts yield significantly higher safety scores compared to fixed system prompts and static self-critique defenses. Code released at https://github.com/vicgalle/meta-self-critique.git .