Critical RCE fixed in Kedro logging configuration
TL;DR — A Critical Kedro startup-path bug can lead to remote code execution by loading an attacker-controlled Python logging configuration via KEDRO_LOGGING_CONFIG.
What happened
Kedro is a Python toolbox for building production-ready data science and ML pipelines. CVE-2026-35171 reports a Critical remote code execution (RCE) issue where Kedro loads a logging configuration file specified by the KEDRO_LOGGING_CONFIG environment variable without validation.
Per the CVE record, the issue stems from unsafe use of logging.config.dictConfig() with attacker-influenced configuration: Python logging configuration supports a special () key that enables arbitrary callable instantiation, which can be abused to execute system commands during application startup.
| Item | Source value |
|---|---|
| Affected package | kedro |
| Affected versions | < 1.3.0 |
| Severity | CVSS v3.1 9.8 (Critical) |
| Root cause class | Code injection / unsafe deserialization patterns in config handling |
Who is impacted
- Projects using
kedroversions< 1.3.0. - Environments where an attacker can influence
KEDRO_LOGGING_CONFIG(or otherwise control the logging configuration content loaded at startup), which is a common risk pattern in shared build/runner, notebook, or multi-tenant execution contexts.
What to do now
- Follow vendor remediation guidance and apply the latest patched release available at the time of writing.
"This vulnerability is fixed in 1.3.0."
- Inventory where
kedrois present in production and automation (lockfiles, SBOM/SCA results, container images, scheduled pipeline jobs) and prioritize anything that starts Kedro based on environment variables. - Audit how
KEDRO_LOGGING_CONFIGis set and propagated (CI variables,.envfiles, container env, orchestration manifests) and reduce who/what can modify it. - If compromise is suspected, treat this as a startup-time code execution path: review recent runs for unexpected logging config paths and rotate credentials accessible to the affected runtime.
Content is AI-assisted and reviewed by our team, but issues may be missed and best practices evolve rapidly, send corrections to [email protected]. Always consult official documentation and validate key implementation decisions before making design or security choices.