CVE-2026-111391: Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection') in sgl-project sglang

CVE-2026-111391 describes an improper neutralization of directives in dynamically evaluated code ('Eval Injection') vulnerability in sgl-project's SGLang product. Specifically, versions 0.5.10 through 0.5.22 have a flaw in the _evaluate_filter() function that allows unauthenticated remote attackers to execute arbitrary Python code. This is achieved by escaping the eval() sandbox using re.search __globals__ and sending a crafted filter expression to the /dumper/configure endpoint when the environment variable DUMPER_SERVER_PORT is set. The malicious code runs in the scheduler process on the next forward pass. The CVSS 4.0 score is 9.2, indicating critical severity with network attack vector, low attack complexity, no privileges or user interaction required, and high impact on confidentiality, integrity, and availability.

Successful exploitation allows unauthenticated remote attackers to execute arbitrary Python code on the affected system, leading to full compromise of the scheduler process. This can result in complete control over the affected system, data theft, service disruption, or further lateral movement. The vulnerability is critical due to its remote, unauthenticated nature and high impact on system security.

No patch or official fix information is provided in the available data. Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, restrict access to the /dumper/configure endpoint and avoid setting the DUMPER_SERVER_PORT environment variable if possible to reduce exposure.