CVE-2026-75062

Eval Injection in google/langfun via default lf.query protocol

Published Aug 26, 2026 CVSS 4.0 · 9.2 Critical EPSS < 1% KEV · Not listed
Description
Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection') in the default lf.query Python protocol in Google langfun versions prior to 0.1.2 allows remote unauthenticated attackers to execute arbitrary Python code in the context of the host application via crafted prompt inputs that cause the model to generate executable Python expressions evaluated without a sandbox.

Analysis

AI-generated

Technical impact

The vulnerability is an eval injection flaw in the default lf.query Python protocol of Google langfun. An attacker who can send a prompt to the model can cause the system to generate a Python expression that is then evaluated with the process’s full permissions. This flaw provides remote unauthenticated execution of arbitrary Python code, classified under CWE-1188 (Improper Neutralization of Directives) and CWE-95 (Improper Neutralization of Code).

Affected Systems

Google langfun versions earlier than 0.1.2 that expose the default lf.query protocol are affected. Any deployment that has enabled this protocol and is still running a pre‑0.1.2 release is at risk.

Risk and Exploitability

The CVSS score of 9.2 marks this issue as critical. The EPSS score is not provided, so the current exploitation likelihood cannot be quantified. The bug is exploitable remotely without any authentication, simply by sending a crafted prompt that prompts the model to return code evaluable by the host. Because the evaluation occurs without a sandbox, the attacker can gain control of the host process and potentially compromise confidentiality, integrity, and availability.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade Google langfun to version 0.1.2 or later to apply the patch that removes unsafe eval usage.
  • If upgrading is not possible, disable or remove the default lf.query protocol from the application so that no dynamic code evaluation is performed.
  • Consider implementing an additional sandbox or execution guard around any remaining eval-like functionality to prevent untrusted code from accessing critical resources.

Generated by OpenCVE AI on August 26, 2026 at 17:06 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
First Time appeared Google
Google langfun
Vendors & Products Google
Google langfun
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'poc', 'Technical Impact': 'total'}, 'version': '2.0.3'}

Type Values Removed Values Added
Description Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection') in the default lf.query Python protocol in Google langfun versions prior to 0.1.2 allows remote unauthenticated attackers to execute arbitrary Python code in the context of the host application via crafted prompt inputs that cause the model to generate executable Python expressions evaluated without a sandbox.
Title Eval Injection in google/langfun via default lf.query protocol
Weaknesses CWE-1188
CWE-95
References
Metrics cvssV4_0

{'score': 9.2, 'vector': 'CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N'}

Weaknesses
  • CWE-1188

    Initialization of a Resource with an Insecure Default

  • CWE-95

    Improper Neutralization of Directives in Dynamically Evaluated Code ('Eval Injection')

Sources

Status
PUBLISHED
Assigner
Google
Published
Updated
Aug 27, 2026, 4:55 PM
Reserved
Aug 17, 2026, 3:35 PM
Status
Awaiting Analysis
Published
Aug 26, 2026, 3:16 PM
Modified
Aug 31, 2026, 6:50 PM