MemAudit is a new framework designed to audit the memory of large language model agents to identify and mitigate the effects of malicious memory injections. The framework has demonstrated a significant reduction in attack success rates, indicating a promising advancement in the security of AI systems that utilize persistent memory.