Jerry Wei

The paper introduces Trojan-Speak, an adversarial fine-tuning method that successfully bypasses advanced LLM safety classifiers (like Anthropic's Constitutional Classifiers) with minimal degradation to the model's core reasoning capabilities.

The paper introduces a robust streaming probing objective that requires multiple evidence tokens to support a prediction, significantly improving the detection of harmful intent in LLMs, especially in sensitive CBRN domains.

The paper demonstrates that advanced jailbreaks do not impose a significant 'jailbreak tax' on highly capable frontier language models, retaining near-native performance.