← All stories
newsAI Agents (OpenClaw, Grok Bot, Hermes, Etc.) · Nov 10, 2026

Grok Bot security report: unauthenticated Docker API, hard-coded token, plaintext cookies, no fix since August

Security firm Profero published a write-up on November 8 describing serious flaws in the Grok Bot agent sandbox, including an unauthenticated Docker API leading to root on the virtual machine, a hard-coded local auth token, browser cookies stored in plain text, an open Chrome debugging port, and every agent sharing the same Linux user. Profero says it reported the issues to xAI in August, with no fix, advisory or CVE so far. The discussion asks whether users should change how they use Grok Bot or other agent platforms.

Read the full source →