Abstract
Concept erasure aims to remove semantic concepts from a trained generative model and is increasingly important for responsible AI deployment. However, verifying whether a model has robustly removed targeted concepts remains a critical challenge. Existing evaluation methods are typically pre-defined and static, failing to expose vulnerabilities under diverse natural-language probes and challenging conditions. Moreover, manually designed evaluation strategies can be biased and difficult to scale.
We posit that concept erasure evaluation is best formulated as an adaptive hypothesis search, operationalised by agents that iteratively propose, critique, and verify tests to systematically expand coverage of failure modes. To this end, we propose Stress Testing Agents for Concept Erasure (STACE), a framework that autonomously stress-tests concept-erased models using multiple Large Language Model (LLM) agents, by iteratively generating and verifying stress-testing hypotheses grounded by external knowledge.
We also introduce a suite of metrics for assessing the performance and efficiency of LLM-agent-powered stress-testing frameworks. Our extensive experiments show that STACE outperforms five LLM-based evaluation baselines on four concept categories. Further analysis across two T2I models, six concept erasure approaches, and various erasure strengths shows that STACE is robust for different settings. We also show that STACE can be adapted beyond concept erasure evaluation to other problem domains, such as LLM jailbreaking.
Our code is available anonymously.
Blogger's Review: This research advances the verification methods of concept erasure by introducing the STACE framework, emphasizing the importance of adaptive hypothesis search. This innovative approach enhances evaluation flexibility and effectiveness while providing new insights for future AI safety research. Its adaptability across various models and methods showcases its broad application potential.