CVE-2026-115885: CWE-532: Insertion of Sensitive Information into Log File in langchain-ai langgraph

The LangGraph CLI (langgraph) versions >=0.4.15 and <0.4.32 have a CWE-532 vulnerability where the 'langgraph deploy --verbose' command passes a temporary registry push token via standard input to 'docker login --password-stdin'. However, the CLI subprocess helper prints this command input in verbose mode, causing the token to be logged. This exposure allows an attacker who can read CI logs or terminal recordings to use the token to pull or push images to the deployment's image repository during the token's maximum 12-hour lifetime. The token alone cannot update deployments to execute pushed images. Deployments without verbose mode and remote image builds are not affected. The vulnerability is resolved in langgraph version 0.4.32.

Exposure of temporary registry push tokens in logs or terminal recordings can lead to unauthorized image repository access, allowing attackers to pull or push container images within the token's validity period (up to 12 hours). This could compromise the integrity of deployment images. However, the token does not permit updating deployments to execute pushed images, limiting the scope of impact. Deployments not using verbose mode or remote image builds are unaffected.

Upgrade langgraph to version 0.4.32 or later, where this logging of sensitive tokens in verbose mode is fixed. Avoid using the '--verbose' flag during deployment until the update is applied. No other mitigation is indicated by the vendor advisory.