Product
LangChain is vulnerable to unsafe deserialization of attacker-controlled objects through overly broad `load()` allowlists, potentially leading to persistent chat-history poisoning, prompt injection, credential disclosure, or server-side requests.