Project Zero is exploring the potential of Large Language Models (LLMs) in vulnerability research. Despite initial low scores in the CyberSecEval2 benchmark, refined testing methodologies can significantly improve LLM performance. Project Zero proposes guiding principles for evaluating LLMs, focusing on providing ample space for reasoning, addressing model limitations, and ensuring realistic testing scenarios. Implementing these principles in their framework increased CyberSecEval2 performance, achieving top scores on Buffer Overflow tests and improved results on Advanced Memory Corruption tests. While progress has been made, Project Zero emphasizes the need for more challenging benchmarks and effective methodologies to fully leverage LLM capabilities.