Hermes WebUI Unauthenticated Terminal RCE

Technical Analysis

CVE-2026-58123 stems from missing authentication checks on Hermes WebUI's embedded terminal endpoints, which are designed to provide developers or operators with browser-based shell access. An attacker can sequentially call the session-creation, PTY-attachment, and command-write endpoints to obtain full command execution as the server process user, with no credentials or prior access required. This is a classic unauthenticated command injection/RCE pattern common in web-based admin consoles and IDE-style tools that embed live terminal functionality for convenience. Because Hermes WebUI is frequently bundled into developer tooling, notebook environments, and orchestration dashboards, hosts running it may also host or interface with AI agent runtimes, RAG pipeline services, or LLM tool-execution backends, meaning a successful exploit could grant attackers direct control over agent infrastructure, exfiltrate API keys/secrets used by agents, or pivot into connected model-serving environments.

Affected Systems

Hermes WebUI versions prior to 0.51.788, including any deployment exposing the embedded terminal API endpoints (session creation, PTY attach, terminal input) over HTTP without authentication middleware.

Indicators of Compromise

• Endpoint pattern: /api/terminal/session (unauthenticated POST)
• Endpoint pattern: /api/terminal/attach
• Endpoint pattern: /api/terminal/input
• Anomalous sequential unauthenticated requests to terminal API paths within short time windows
• Unexpected child shell processes spawned by Hermes WebUI server process

Remediation Steps

1. 1

   Upgrade Immediately

   Update Hermes WebUI to version 0.51.788 or later, which patches the authentication bypass on terminal API endpoints.

2. 2

   Restrict Network Exposure

   Ensure Hermes WebUI instances are not exposed to the public internet; restrict access to trusted internal networks or VPN-only access via firewall rules.

3. 3

   Enforce Authentication at Proxy Layer

   If immediate patching is not possible, place a reverse proxy in front of Hermes WebUI that enforces authentication and blocks direct access to /api/terminal/* endpoints.

4. 4

   Audit Logs for Exploitation

   Review server access logs for unauthenticated calls to terminal session/attach/input endpoints and inspect process trees for unexpected shell execution.

5. 5

   Rotate Credentials and Agent API Keys

   If exposure is confirmed or suspected, rotate all API keys, tokens, and credentials accessible from the compromised host, especially those used by connected AI agent or RAG services.

6. 6

   Network Segmentation

   Isolate hosts running Hermes WebUI from sensitive agent orchestration, model-serving, or secrets-management infrastructure to limit lateral movement.