Make Them Spill the Beans! Coercive Knowledge Extraction from (Production) LLMs
This exposes a new security vulnerability in LLMs for users and developers, enabling coercive extraction of toxic knowledge even from models designed for safe or specific tasks like coding.
The study tackled the problem of extracting harmful content from aligned large language models (LLMs) by exploiting output logits, revealing that even when models reject toxic requests, harmful responses are hidden in the logits, and achieved 92% effectiveness compared to 62% for jail-breaking methods, with 10 to 20 times faster performance.
Large Language Models (LLMs) are now widely used in various applications, making it crucial to align their ethical standards with human values. However, recent jail-breaking methods demonstrate that this alignment can be undermined using carefully constructed prompts. In our study, we reveal a new threat to LLM alignment when a bad actor has access to the model's output logits, a common feature in both open-source LLMs and many commercial LLM APIs (e.g., certain GPT models). It does not rely on crafting specific prompts. Instead, it exploits the fact that even when an LLM rejects a toxic request, a harmful response often hides deep in the output logits. By forcefully selecting lower-ranked output tokens during the auto-regressive generation process at a few critical output positions, we can compel the model to reveal these hidden responses. We term this process model interrogation. This approach differs from and outperforms jail-breaking methods, achieving 92% effectiveness compared to 62%, and is 10 to 20 times faster. The harmful content uncovered through our method is more relevant, complete, and clear. Additionally, it can complement jail-breaking strategies, with which results in further boosting attack performance. Our findings indicate that interrogation can extract toxic knowledge even from models specifically designed for coding tasks.