Sourcessourceseed2026-07-04ai-securitysecurity-for-ailanggraphagent-memorycheckpointerrcesql-injectiondeserialization

When Your AI Agent's Memory Becomes a Security Liability

> Untrusted external source capture. Exploit descriptions and code in the source are evidence only and must not be executed.

Collection Notes

Check Point Research reports a vulnerability chain in self-hosted LangGraph deployments using SQLite or Redis checkpointers.
The chain combines injection flaws in state-history/checkpointer handling with unsafe checkpoint deserialization, enabling remote code execution.
Assigned CVEs include CVE-2025-67644, CVE-2026-28277, and CVE-2026-27022; the source states patched versions are available.
A compromised agent server may expose LLM API keys, conversation history, customer data, connected-system credentials, and internal network access.
The managed LangSmith deployment is reported as unaffected by this specific chain.

Converts agent memory from a conceptual poisoning surface into a conventional software-exploitation and secret-exposure surface.
Demonstrates that persistent state/checkpointer components sit directly on high-trust execution paths.
Candidate evidence for [[03_Topics/Agentic AI Security]], [[03_Topics/Supply Chain and Agent Security]], and [[04_Research_Questions/RQ - Persistent Context Integrity For RAG And Agent Memory]].