Security
DeepSeek Harness Flaw Let AI Agents Disable Their Own File Sandbox Without Approval
The Hacker News·September 9, 2026·1 min read
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. The agent could remove the operating-system limit that prevents writing outside its workspace, without approval, by calling the tool's own web interface.
Read at The Hacker NewsDaily crypto arcade
Read the news, then play it.
Chainshorts turns crypto headlines into a daily game. Catch up in 60 words, then jump into daily lucky draws for a shot at the pot.