Brief

Researchers have developed Reflect-Guard, a new method to improve the safety of large language models against adversarial prompts. This technique uses chain-of-thought self-reflection, fine-tuning models like Llama-Guard-3-8B with distilled reasoning from GPT-4o-mini. Even with a small dataset and minimal parameter updates, Reflect-Guard significantly boosts performance on benchmarks designed to test defenses against jailbreak attacks, particularly by enabling models to reason through obfuscated malicious intent. AI

Researchers have developed LASH, a novel framework designed to enhance the jailbreaking of large language models. LASH adaptively combines outputs from multiple existing attack methods, treating them as seed prompts. This approach leverages the complementary strengths of different attack families to improve success rates against various models and harm categories. In evaluations on the JailbreakBench dataset, LASH achieved high attack success rates with significantly fewer queries compared to state-of-the-art baselines. AI