Researchers describe hiding payloads in open-weight AI models
The Lazarus method can use a tiny share of neurons; hiding payloads in weights is described as research, not a known campaign.
8 stories, newest first · one card per story
The Lazarus method can use a tiny share of neurons; hiding payloads in weights is described as research, not a known campaign.
A user must open a malicious page or file; ZDI rates CVE-2026-118623 at CVSS 7.8.

Guidance covers 30 risk types, unsigned model files, and backdoors that behavioral tests may miss.

A patched Google DevSite flaw let unauthenticated attackers forge Flask sessions and achieve remote code execution.

JFrog released a proof-of-concept for unauthenticated remote code execution in LMCache, tracked as CVE-2026-105192, with no patch.

Unpatched LMCache flaw CVE-2026-105192 allows unauthenticated remote code execution, and JFrog published a PoC.

Unpatched LMCache flaw CVE-2026-105192 lets unauthenticated attackers run code via pickle over ZeroMQ.

Open-weight AI models can hide backdoors and poisoning that scanners miss, shifting breach liability onto security leaders.
Today 1K readers · 1.2K views · 174 reads · all time 202K views