vulnerability ·
LangChain serialization can expose secrets through untrusted agent data
Untrusted metadata can survive LangChain serialization as object instructions and expose environment secrets when vulnerable deserialization settings are used.
critical severityconfirmed confidencepatchedNo known exploitation
Quick answer
What defenders need to know
- What is it?
- Untrusted metadata can survive LangChain serialization as object instructions and expose environment secrets when vulnerable deserialization settings are used.
- Who is affected?
- langchain-core langchain-core < 0.3.81, langchain-core >= 1.0.0, < 1.2.5
- What should defenders do?
- Upgrade langchain-core to 0.3.81 on the 0.3 branch, or 1.2.5 on the 1.x branch; prefer a later supported release.
- Is a fix available?
- Yes. Upgrade to langchain-core 0.3.81, langchain-core 1.2.5.
01 / Response
What to do
- Upgrade langchain-core to 0.3.81 on the 0.3 branch, or 1.2.5 on the 1.x branch; prefer a later supported release.
- Keep environment-secret loading disabled and restrict deserialization to required object types.
02 / Technical assessment
Evidence-backed analysis
The affected paths include streaming, message history and loading untrusted vector-store documents. Risk depends on processing attacker-controlled data, not simply installing LangChain.
- Official advisory identifies unsafe serialization and environment-secret exposure.
- No verified in-the-wild activity is established by this record.
03 / Timeline
Disclosure timeline
- Advisory published
Maintainer disclosed GHSA-c67j-w6g6-q2cm.
- Advisory revised
Maintainer updated the source advisory.
04 / Standards crosswalk
Standards crosswalk
FrameworkControl IDStatusEvidence
Control ID
LLM02risk alignmentStatusCandidate mapping
EvidenceNo supporting evidence published
Control ID
LLM01risk alignmentStatusCandidate mapping
EvidenceNo supporting evidence published
05 / Primary sources