Skip to main content
CVE Vulnerability Database
Vulnerability Database/CVE-2026-14626

CVE-2026-14626: NousResearch Hermes-Agent DoS Vulnerability

CVE-2026-14626 is a denial of service flaw in NousResearch hermes-agent affecting the HTTP API component. Attackers can remotely exploit this vulnerability via the todos argument. This article covers technical details, affected versions, impact assessment, and mitigation strategies.

Published:

CVE-2026-14626 Overview

CVE-2026-14626 is a denial of service vulnerability affecting NousResearch hermes-agent versions up to 2026.4.30. The flaw resides in the AIAgent.run_conversation function within run_agent.py, which is part of the HTTP API component. Attackers can manipulate the todos argument to trigger resource exhaustion and disrupt service availability. The attack is remotely exploitable and requires low-level privileges. A public exploit has been released, increasing the likelihood of opportunistic attacks. The vendor was contacted prior to public disclosure but did not respond, leaving affected deployments without official remediation guidance.

Critical Impact

Remote attackers with low privileges can send crafted todos payloads to the HTTP API to induce denial of service in the hermes-agent runtime.

Affected Products

  • NousResearch hermes-agent versions up to and including 2026.4.30
  • Deployments exposing the hermes-agent HTTP API to untrusted networks
  • Integrations invoking AIAgent.run_conversation in run_agent.py

Discovery Timeline

  • 2026-07-04 - CVE-2026-14626 published to NVD
  • 2026-07-07 - Last updated in NVD database

Technical Details for CVE-2026-14626

Vulnerability Analysis

The vulnerability is categorized under [CWE-404] Improper Resource Shutdown or Release. The affected function AIAgent.run_conversation in run_agent.py processes the todos argument received through the HTTP API without adequate constraints on resource consumption. An authenticated remote attacker can submit a manipulated todos parameter that causes the agent to exhaust resources or fail to release them, resulting in denial of service. Because hermes-agent is an AI agent framework, disruption of the runtime affects downstream automation and conversational workflows relying on the service.

Root Cause

The root cause is improper handling of the todos argument within the conversation loop. The function does not enforce sufficient bounds or lifecycle controls on task-related resources. When an attacker supplies crafted input, the agent enters a state where it cannot properly reclaim or terminate resources, degrading availability.

Attack Vector

The attack is initiated over the network against the HTTP API. The attacker requires low-level authentication but no user interaction. A public proof-of-concept is available on GitHub Gist, published by the reporter. Technical details are described in the GitHub Gist PoC and the VulDB CVE-2026-14626 entry.

Detection Methods for CVE-2026-14626

Indicators of Compromise

  • Unusually large or malformed todos payloads in HTTP requests to the hermes-agent API
  • Sudden spikes in CPU, memory, or file descriptor usage on hosts running run_agent.py
  • Repeated AIAgent.run_conversation invocations from a single source followed by service unresponsiveness

Detection Strategies

  • Instrument the HTTP API to log request sizes and structure of the todos argument for offline analysis
  • Baseline normal request patterns to hermes-agent endpoints and alert on statistical deviations
  • Correlate authentication events with abnormal request volumes to identify low-privilege abuse

Monitoring Recommendations

  • Monitor process-level metrics for the hermes-agent runtime and generate alerts on sustained resource pressure
  • Track HTTP 5xx response rates and request latency on the API endpoint
  • Review authentication logs for accounts making disproportionate calls to conversation endpoints

How to Mitigate CVE-2026-14626

Immediate Actions Required

  • Restrict network exposure of the hermes-agent HTTP API to trusted management networks only
  • Enforce input size limits and schema validation on the todos argument at an upstream proxy
  • Rotate and scope credentials used by clients of hermes-agent to reduce blast radius
  • Deploy rate limiting on the AIAgent.run_conversation endpoint

Patch Information

No vendor patch is available. According to the disclosure, NousResearch did not respond to the reporter. Operators should track the VulDB Vulnerability #376142 record for future updates and monitor the upstream project for new releases beyond version 2026.4.30.

Workarounds

  • Place hermes-agent behind a reverse proxy that enforces request body size and JSON depth limits
  • Configure a Web Application Firewall rule to reject requests with anomalous todos structures
  • Run hermes-agent under strict cgroup or container resource limits to contain exhaustion attempts
  • Disable public exposure of the HTTP API until a fixed release is available
bash
# Example reverse proxy limits (nginx)
client_max_body_size 64k;
limit_req_zone $binary_remote_addr zone=hermes:10m rate=10r/m;
location /run_conversation {
    limit_req zone=hermes burst=5 nodelay;
    proxy_pass http://hermes_agent_upstream;
}

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Default Legacy - Prefooter | Experience the World’s Most Advanced Cybersecurity Platform

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.