Security brief

ShadowMQ research traced unsafe deserialization across AI inference projects

REVIEW

Oligo’s findings concerned exposed messaging interfaces and Python deserialization, with different advisories for individual projects.

Published 29 Oct 2026 · Material update 14 Nov 2025 — last time the underlying evidence changed · Evidence: medium

What happened

November 2025 reporting described Oligo Security’s ShadowMQ research into unsafe deserialization in AI inference systems. The recurring pattern combined network-accessible ZeroMQ interfaces with Python pickle-based object handling.

The report traced related implementations across multiple projects. Shared code patterns do not make every listed product equally exposed or give them one universal affected-version range.

Inventory inference services and restrict internal messaging endpoints to intended peers. Consult each project’s advisory for its affected versions and fix; an inference endpoint should not accept untrusted serialized objects simply because it sits on an internal network.

Read the original source.

Next step

Restrict inference messaging interfaces and apply each project’s relevant security fixes.

What remains unknown

Open question
The sources reviewed do not establish whether your deployment was exploited.

Source references