CVE-2026-37003

Remote Code Execution through Unfiltered LLM Input in Agno PythonTools

Published Aug 27, 2026 CVSS 3.1 · 9.8 Critical EPSS 1.3% KEV · Not listed
Description
Agno up to and including 2.5.8 is vulnerable to Remote Code Execution (RCE) via prompt injection. The PythonTools and ShellTools components pass unsanitized, LLM-generated arguments directly to execution sinks including exec(), runpy.run_path(), and subprocess.run(). An unauthenticated attacker can exploit this by embedding malicious instructions in content processed by the agent (such as web pages or documents), allowing for arbitrary code and OS command execution on the host server.

Analysis

AI-generated

Technical impact

Agno software up to and including version 2.5.8 is vulnerable to remote code execution because the PythonTools and ShellTools components accept large language model–generated arguments and pass them directly to native execution functions such as exec(), runpy.run_path(), and subprocess.run() without performing any sanitization. A malicious user who can embed commands into content processed by the agent—such as web pages or documents—can therefore cause arbitrary Python code and operating‑system command execution on the host server, leading to complete compromise of confidentiality, integrity, and availability.

Affected Systems

The affected product is Agno, with all releases up to and including 2.5.8 being impacted. No other vendors or product versions are listed as affected.

Risk and Exploitability

The vulnerability requires no authentication and is exploitable via content that the agent processes, which can be supplied by remote entities, making it potentially attractive to attackers. The CVSS score of 9.8 indicates a high severity, while the EPSS score of less than 1 % suggests a low but nonzero likelihood of exploitation. Because the issue is not listed in CISA’s KEV catalog, it may have slipped past many monitoring efforts, but the combination of a high CVSS and the absence of authentication makes it a very concerning risk for exposed deployments.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Apply the vendor‑issued patch or upgrade Agno to a version newer than 2.5.8 as soon as it becomes available.
  • Disallow direct use of exec(), runpy.run_path(), and subprocess.run() for any LLM‑generated input within the PythonTools and ShellTools components.
  • Sanitize or validate all LLM‑generated arguments before they reach execution sinks.
  • Run the Agno process inside a sandbox or container and monitor the host for unexpected command execution.

Generated by OpenCVE AI on September 2, 2026 at 08:52 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
Title Remote Code Execution through Unfiltered LLM Input in Agno PythonTools
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'yes', 'Exploitation': 'none', 'Technical Impact': 'total'}, 'version': '2.0.3'}

Type Values Removed Values Added
Title Remote Code Execution via Unsanitized Prompt Injection in Agno 2.5.8
Weaknesses CWE-78
Type Values Removed Values Added
Metrics cvssV3_1

{'score': 9.8, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H'}

Type Values Removed Values Added
First Time appeared Agno
Agno agno
Vendors & Products Agno
Agno agno
Type Values Removed Values Added
Title Remote Code Execution via Unsanitized Prompt Injection in Agno 2.5.8
Weaknesses CWE-78
CWE-94
Type Values Removed Values Added
Description Agno up to and including 2.5.8 is vulnerable to Remote Code Execution (RCE) via prompt injection. The PythonTools and ShellTools components pass unsanitized, LLM-generated arguments directly to execution sinks including exec(), runpy.run_path(), and subprocess.run(). An unauthenticated attacker can exploit this by embedding malicious instructions in content processed by the agent (such as web pages or documents), allowing for arbitrary code and OS command execution on the host server.
References
Weaknesses
  • CWE-94

    Improper Control of Generation of Code ('Code Injection')

Sources

Status
PUBLISHED
Assigner
mitre
Published
Updated
Sep 1, 2026, 7:34 PM
Reserved
Apr 6, 2026, 12:00 AM
Status
Deferred
Published
Aug 27, 2026, 8:17 PM
Modified
Sep 8, 2026, 7:29 PM