CVE-2026-123675: OpenBMB Internet of Agents code injection in Code Executor Tool

CVE-2026-123675 · Severity: high · CVSS 8.3 · Published 2026-11-06

OpenBMB Internet of Agents is a framework for building collaborative AI agents. An attacker can inject arbitrary code through the Code Executor Tool, which claims to run code in isolated containers but actually executes it directly on the host system with the agent's privileges. A malicious agent joining a team can trick the victim agent into executing arbitrary Python, Bash, or other shell commands, or writing files anywhere on the system.

Technologies: OpenBMB Internet of Agents. Vendors: OpenBMB.

Quick answers

Is CVE-2026-123675 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-123675 being exploited in the wild.
How severe is CVE-2026-123675?
High severity, with a CVSS score of 8.3, published 6 November 2026.
Which products and versions are affected?
OpenBMB Internet of Agents up to commit 9d78e88e47160eb42ee1ee8538c05e1c2c47877a.
Is there a fix for CVE-2026-123675?
No fixed version is recorded yet. Restrict agent collaboration to trusted participants only. Disable or remove the Code Executor Tool if not required. Deploy agents with minimal file system permissions and network isolation at the container/orchestration level. 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-2026-123675 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-123675?
Check whether you run OpenBMB Internet of Agents 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.

How to fix

Restrict agent collaboration to trusted participants only. Disable or remove the Code Executor Tool if not required. Deploy agents with minimal file system permissions and network isolation at the container/orchestration level.

Technical details

The execute_code function in code_executor.py contains multiple vulnerabilities: unsanitized filename arguments allow path traversal (via ../ sequences) to write files outside intended directories, language arguments directly map to local shells without restriction, and code executes in an unsandboxed subprocess with no memory, CPU, or process limits enforced. The vulnerability requires a malicious agent to be present in the collaboration team to trigger the vulnerable tool call through the victim's LLM.

Affected products

  • OpenBMB Internet of Agents up to commit 9d78e88e47160eb42ee1ee8538c05e1c2c47877a

Timeline

  • 2026-10-14: disclosed: Security issue reported on GitHub
  • 2026-11-06: advisory: CVE-2026-123675 published

References

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