Critical Langchain Vulnerability Let attackers Exfiltrate Sensitive Secrets from AI systems

A critical deserialization vulnerability (CVE-2025-68664) in LangChain-core's dumps()/dumpd() handling of user-controlled dictionaries allowed attackers to cause deserialization of untrusted data (leading to environment variable exfiltration and potential RCE via SSRF or Jinja2 rendering). The flaw—assigned CVSS 9.3—was reported in December 2025, patched in langchain-core 0.3.81 and 1.2.5, and mitigations include upgrading, disabling secret resolution by default, auditing serialization/deserialization in streaming/logging/caching, and treating LLM outputs as untrusted.