Skip to content
← Blog · · df00tech

CVE-2026-35002: Arbitrary Code Execution via eval() in Agno's FunctionCall field_type (CVSS 9.8)

vuln-intel pip CVE-2026-35002 #checker-reject

What's Happening

CVE-2026-35002 is a critical arbitrary code execution vulnerability in the Agno AI agent framework for Python (pip package agno). The root cause (CWE-95, Eval Injection) is straightforward but severe: the field_type parameter on a FunctionCall input field is passed directly to Python's eval(). Any attacker who can influence that value can execute arbitrary Python code in the context of the Agno process — no sandboxing, no type validation, just raw code execution.

Affected Software

  • Package: agno (PyPI)
  • Vendor: pip / PyPI ecosystem
  • Affected versions: prior to 2.3.24
  • CVSS: 9.8 (Critical)

No official patch date is recorded at this time; organizations running Agno-based agents should treat any pre-2.3.24 deployment as vulnerable until confirmed otherwise.

Exploitation Status

A public proof-of-concept exists (poc-public). This is not yet confirmed as weaponized in the wild or added to CISA's KEV catalog, but public PoC availability for a CVSS 9.8 eval-injection bug significantly lowers the bar for opportunistic exploitation. Defenders should prioritize detection and patching now rather than waiting for in-the-wild confirmation.

How Our Detection Catches It

This detection combines two layers of coverage:

  • Vulnerable installation discovery — surfacing hosts and environments running agno versions below 2.3.24 so defenders can prioritize patching before exploitation occurs.
  • Behavioral exploitation signatures — successful abuse of the eval() sink typically manifests as a Python interpreter process spawning unexpected child processes (shells, curl/wget, or reverse-shell payloads), along with anomalous outbound network connections or unexpected file writes originating from the Agno process tree. The logic focuses on this parent-child process anomaly — Python unexpectedly forking shell or network utilities — as the primary high-fidelity signal of a triggered eval injection, supplemented by unusual network/file activity correlated to the same process.

We ship this logic across Microsoft Sentinel (KQL), Splunk (SPL), Elastic (EQL), IBM QRadar (AQL), Sumo Logic, Google Chronicle (YARA-L), and CrowdStrike (CQL), so SOC teams can deploy consistent detection regardless of their SIEM/EDR stack.

Next Steps

Full detection logic, sample events, investigation guidance, and atomic test procedures for CVE-2026-35002 are available on the detection detail page. Patch to Agno 2.3.24 or later, and deploy the provided rules across your monitored platforms to catch exploitation attempts in the interim.

Get new detections in your inbox

New ATT&CK coverage plus CISA KEV / CVE detection rules, roughly weekly. No spam, unsubscribe anytime.