CVE-2026-14628: NousResearch hermes-agent Path Traversal via Live Webhook Endpoint

Technical summary

CVE-2026-14628 is a CWE-22 path traversal vulnerability in the extract_media function within gateway/platforms/base.py of NousResearch hermes-agent. The Live Webhook Endpoint fails to properly validate or sanitize file path inputs, enabling directory traversal attacks. The vulnerability requires no authentication, no user interaction, and can be triggered over the network. With a CVSS 3.1 base score of 5.3 (Medium), the attack vector is network-based with low complexity, resulting in limited confidentiality impact but no integrity or availability compromise.

Business impact

This vulnerability enables unauthorized information disclosure through remote path traversal attacks. Organizations deploying hermes-agent may have sensitive operational data, API keys, database credentials, or application source code exposed if an attacker exploits this weakness. The public availability of exploit code significantly increases risk, as opportunistic attackers can quickly weaponize the vulnerability. However, the Medium severity rating indicates the exposure is informational rather than enabling full system compromise or service disruption.

Affected systems

NousResearch hermes-agent versions up to and including 2026.5.16 are vulnerable. The affected component is the Live Webhook Endpoint in gateway/platforms/base.py. Any deployment of hermes-agent prior to 2026.5.17 or later should be considered at risk. Organizations should verify their exact version against the vendor advisory and patch accordingly.

Exploitability

The vulnerability is readily exploitable over the network without authentication or user interaction required. Public exploit code is already available, dramatically lowering the barrier to exploitation. An attacker with network access to the Live Webhook Endpoint can craft path traversal payloads to read arbitrary files the application process has access to. The low attack complexity means standard exploitation tools or simple HTTP requests are sufficient.

Remediation

Upgrade NousResearch hermes-agent to a patched version beyond 2026.5.16 as soon as possible. Verify the exact patched version against the vendor advisory. If immediate patching is not feasible, implement network segmentation and access controls to limit exposure of the Live Webhook Endpoint to trusted internal networks only. Monitor webhook traffic for suspicious path patterns such as dot-dot-slash sequences or encoded traversal sequences.

Patch guidance

Contact NousResearch or check their official repository and security advisory channels for patched versions beyond 2026.5.16. Verify patch availability before deployment planning. Note that the vendor did not respond to early disclosure outreach, so updates may be delayed or require community-driven monitoring. Test patches in a non-production environment first to ensure compatibility with your webhook integrations and downstream services.

Detection guidance

Monitor HTTP requests to the Live Webhook Endpoint for path traversal indicators: patterns containing '../', '..%2f', '%2e%2e%2f', or similar URL-encoded variants. Log all extract_media function calls and file access attempts originating from the webhook component. Check web server and application logs for 400-level errors combined with path traversal attempts. Set up alerts for any requests attempting to access system directories (/etc, /var, Windows\ equivalents) through the webhook interface.

Why prioritize this

Although classified as Medium severity, this vulnerability warrants prompt attention due to public exploit availability and the informational risk to deployed systems. Organizations running hermes-agent in production should patch within 30 days. Prioritize systems where the Live Webhook Endpoint is internet-facing or exposed to untrusted networks. Those with strict network boundaries limiting access to the endpoint may defer slightly but should still patch on a defined schedule.

Risk score, explained

The CVSS 3.1 score of 5.3 reflects a network-exploitable path traversal with no authentication barriers but limited to confidentiality impact (CWE-22). The Medium severity is appropriate because the attack cannot modify or delete data, nor can it disrupt service availability. However, the score does not fully capture the risk elevation from public exploit code; organizations should consider this a functional upgrade from the base score in practical threat modeling.

Frequently asked questions