Apache Livy EUVD-2025-208637

The CVSS 6.3 score with vector CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:L indicates moderate risk with a favorable attack profile: network-accessible, low attack complexity, and no user interaction required. However, the PR:L (Privileges Required: Low) constraint is significant—an attacker must already have valid credentials to Livy's REST or JDBC interface, reducing the pool of threat actors substantially. The EPSS score is not provided in the available data, but given the required authentication boundary, real-world exploitability is lower than the CVSS number alone suggests. The vulnerability is not listed as actively exploited in known exploitation vectors (KEV status unknown from provided data), and no public proof-of-concept code is mentioned. Under SSVC decision logic: exploitation status is unconfirmed (no KEV), technical impact is moderate (confidentiality, integrity, and availability all affected), and the attack surface is restricted to authenticated users. This should be prioritized for patching in environments where Livy is exposed to multiple users or untrusted internal networks, but is lower priority in isolated, single-user, or air-gapped deployments.

An authenticated user with valid credentials to a Livy instance connected to Spark 3.1+ submits a REST POST request to create a new Spark session, embedding a malicious Spark configuration parameter (e.g., a configuration option introduced in Spark 3.1 that modifies file access semantics or file system principals). Because Livy does not validate this configuration value, it passes it directly to the Spark cluster. The Spark cluster processes the configuration and allows the subsequent file operations requested by the user to access files owned by other users or in restricted directories. The attacker successfully reads sensitive data files they would otherwise not have permission to access. No elevated privileges or social engineering is required—the attacker merely needs an existing Livy account and knowledge of which Spark 3.1+ configuration options can be abused for this purpose.