AI agent memory can quietly store API keys, credentials, and sensitive docs in plain text. It can also be poisoned via plugins, skills, and MCP integrations. Audit agent memory for exposed secrets. #OWASP #Vectorize #MemoryGuard