Agentic AI memory attacks spread across sessions and users, and most organizations aren’t ready

This feed highlights multiple high-risk trends and incidents: researcher Idan Habler disclosed “MemoryTrap,” a memory-poisoning technique that compromised Claude Code’s long-term memory and demonstrated how a single poisoned memory object can propagate across sessions, users, and subagents—underscoring that AI memory should be governed like secrets and identities. Related telemetry shows massive credential sprawl: GitGuardian reported ~28.7M new secrets exposed in public GitHub commits in 2025, amplifying risk for AI agents that rely on many credentials. Active incidents include a CPUID site-s