PYSEC-2026-2272: Ray: Remote Code Execution via Parquet Arrow Extension Type Deserialization
Ray is an AI compute engine. From version 2.54.0 to before version 2.55.0, Ray Data registers custom Arrow extension types (ray.data.arrowtensor, ray.data.arrowtensorv2, ray.data.arrowvariableshapedtensor) globally in PyArrow. When PyArrow reads a Parquet file containing one of these extension types, it calls arrowextdeserialize on the field's metadata bytes. Ray's implementation passes these bytes directly to cloudpickle.loads(), achieving arbitrary code execution during schema parsing, before any row data is read. This issue has been patched in version 2.55.0.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
Rayto a version that resolves this vulnerability.Fixed in 2.55.0
Event History
Frequently Asked Questions
Which Ray deployments are affected?
Ray versions 2.54.0 through versions before 2.55.0 are affected. Version 2.55.0 contains the patch.
What must an attacker provide to trigger exploitation?
An attacker needs a Parquet file containing one of Ray's registered Arrow extension types with malicious field metadata. The vulnerable process must read that file with PyArrow after Ray has globally registered the extension types.
Does processing Parquet row data need to begin for exploitation?
No. The unsafe deserialization occurs during schema parsing, before any row data is read.
What should be done if vulnerable systems may process untrusted Parquet files?
Upgrade Ray to version 2.55.0. Until then, avoid reading untrusted Parquet files in affected Ray environments.