M8M – See what your AI remembers about you and detect memory poisoning (github.com)

The recently announced M8M tool enhances AI memory observability and security by enabling users to monitor what their AI agents remember about them. This tool analyzes memory provenance and detects potential memory poisoning attempts, a growing concern following reports of 50 such incidents across various companies in just two months. M8M offers a comprehensive local SQLite database that logs live memory operations, tracks memory file changes, and utilizes pattern-based scanning to identify sensitive information like credentials and instructive directives. Significantly, more advanced AI models, such as GPT-5.4, have shown increased vulnerabilities to memory injection attacks, with an alarming 87.5% success rate noted. M8M not only provides visibility into an AI's stored knowledge but also empowers users to manage it effectively, allowing for snapshots, rollbacks, and auditing of changes. This tool acts as a "git log" for AI memories, ensuring that every change is recorded, thereby bolstering trust and security in AI systems as they become integral to personal and professional workflows.