CVE-2025-65720: assafelovic GPT Researcher command injection in MCP STDIO

CVE-2025-65720 · Severity: critical · CVSS 9.8 · Published 2026-07-15

GPT Researcher is an autonomous AI agent used to conduct deep research by scraping and analyzing web data. A security flaw in version 3.3.7 allows an attacker to take control of the system running the agent if the user interacts with a specially crafted malicious webpage during a research task. This could lead to the theft of sensitive data, unauthorized access to local files, or a complete compromise of the host machine.

Technologies: Assaf Elovic GPT Researcher. Vendors: Assaf Elovic, PyPI.

Quick answers

Is CVE-2025-65720 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2025-65720 being exploited in the wild.
How severe is CVE-2025-65720?
Critical severity, with a CVSS score of 9.8, published 15 July 2026.
Which products and versions are affected?
Assaf Elovic GPT Researcher 3.3.7; gpt-researcher (PyPI) >= 0.14.5, <= 0.14.7, ECOSYSTEM:introduced=0.14.5,last_affected=0.14.7.
Is there a fix for CVE-2025-65720?
No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.
How is CVE-2025-65720 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2025-65720?
Check whether you run Assaf Elovic GPT Researcher or gpt-researcher (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

Technical details

A command injection vulnerability exists in GPT Researcher v3.3.7, stemming from improper handling of inputs within the Model Context Protocol (MCP) STDIO implementation. The flaw allows an attacker to achieve Remote Code Execution (RCE) by tricking a user into interacting with a crafted HTML page while the autonomous agent is conducting research. The root cause is linked to a systemic issue in Anthropic's MCP SDK where unsanitized inputs are passed to system commands. Attackers can exploit this to execute arbitrary shell commands with the privileges of the application. While the NVD entry lists the severity as 'info', the underlying RCE impact typically warrants a High or Critical rating.

Affected products

  • assafelovic GPT Researcher 3.3.7

Identifiers

  • CVE-2025-65720
  • GHSA-8j86-h8gg-797p
  • PYSEC-2026-4219

Timeline

  • 2026-04-15: disclosed: Initial disclosure by OX Security researchers regarding MCP vulnerabilities.
  • 2026-07-15: advisory: CVE-2025-65720 published to the NVD.

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.