hermes-agent.nousresearch.com Hermes Agent Publishes Defense-in-Depth Security Model and Work-Machine Hardening Guide
Article Content
- •Hermes Agent documents an eight-layer defense-in-depth security model covering authorization, command approval, file writes, and container isolation.
- •Per the guide, dangerous-command approval is enabled by default in smart mode, and unanswered approval prompts time out and are denied.
- •A hardline blocklist refuses certain destructive commands regardless of approval mode, YOLO mode, or allow-always settings, per the documentation.
- •Setting approvals.mode to off or enabling YOLO mode disables or bypasses safety prompts and is advised only for trusted environments.
- •The articles describe design and configuration guidance; they report no vulnerability, breach, or active exploitation.
Nous Research's documentation for Hermes Agent, dated 2026-11-22, describes an eight-layer defense-in-depth security model covering user authorization, dangerous command approval, file write safety, container isolation, MCP credential filtering, context file scanning, cross-session isolation, and input sanitization. A second guide addresses running the agent on a personal or employer-managed workstation. It states that fresh installs have dangerous-command approval on by default (approvals.mode: smart), that approval prompts time out after 300 seconds and deny by default, and that a hardline blocklist refuses certain commands (such as recursive deletes of root, fork bombs, and disk zeroing) regardless of approval mode or YOLO settings. The documentation also notes that approvals.mode: off disables all safety prompts and should be used only in trusted environments, and that YOLO mode bypasses approval prompts for the current session. These are vendor-published configuration and design descriptions; the articles report no incident, vulnerability, or exploitation.
Ask AI about this cluster
Answers cite the sources they use
Timeline
More articles in this cluster (2)
Common questions
Are default installs protected against dangerous commands?
What happens if I don't respond to an approval prompt?
Which settings weaken these protections?
Continue Reading
Critical Apache Roller RCE Vulnerability Actively Exploited in the Wild Apache has issued a warning regarding a critical remote code execution (RCE) vulnerability in Apache Roller, tracked as CVE-2026-129448, which is actively being exploited. This flaw affects versions 4.0.0 through 6.1.4 and allows unauthenticated attackers to execute arbitrary code via the XML-RPC endpoint. The…