Skip to content
AI Sec Watch
MediumVulnerabilityLLM-specific

GHSA-rr7j-v2q5-chgv: LangSmith SDK: Streaming token events bypass output redaction

Source
GitHub Advisory(opens in a new tab)
Identifier
GHSA-rr7j-v2q5-chgv
Published
Record updated
View JSON
Affected
  • langsmith <= 0.7.30, fixed in 0.7.31
  • langsmith <= 0.5.18, fixed in 0.5.19

Summary

The LangSmith SDK's output redaction controls, hideOutputs in JS and hide_outputs in Python, do not cover streaming token events. Each streamed chunk is recorded as a new_token event holding the raw token value, and the redaction pipeline only processes the inputs and outputs fields, never the events array. Applications that rely on output redaction to keep sensitive LLM output out of LangSmith still leak the full streamed content through run events. Both the JS and Python SDKs are affected.

Mitigation

The source does not state a fix yet. Check the original advisory for updates.

Affected packages in the Exposure Registry

Matched by package name and ecosystem. Each entry shows whether the package delegates to a language model and how many tracked packages depend on it.

  • langsmithnpmNo LLM dependency detected · 8 tracked dependents
  • langsmithPyPILLM dependency since 2025-03-14 · 10 tracked dependents

Related items