AnalysisCybersecuritySeptember 9, 2026

DeepSeek Harness flaw let AI agents disable their own sandbox

A flaw in DeepSeek Harness, DeepSeek's open-source tool for running AI coding agents, let a sandboxed agent disable its own sandbox with a single command, tracked as CVE-2026-82533 with a 9.4 severity rating. Fixed on August 27.

1 source

More stories today

Open the live feed