/

CVE-2026-86124

Severity:

CRITICAL RISK

(9.8)

Missing Authentication (CWE-306) in AutoAgent TCP Server

CVE-2026-86124

Deferred

Vulnerability type

Category:

Missing Authentication

CWE:

CWE-306

Impact:

Unauthenticated Remote Code Execution

Severity:

CRITICAL RISK

(9.8)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

AutoAgent contains a critical missing-authentication flaw in its TCP server, which may accept connections from unauthenticated remote clients and execute supplied commands with root privileges. When the server is reachable and insufficiently isolated, an attacker could gain control of the container and access host workspace directories exposed through bind mounts.

Affected context

The vulnerability occurs in AutoAgent deployments where the TCP communication server listens on all network interfaces and exposes its command-execution functionality without an authentication or authorization check. Risk is increased when the communication port is reachable from untrusted networks, the process runs as root, or the container has bind-mounted access to host workspace directories.

Exploit preconditions

  • Attack surface: AutoAgent's TCP communication server
  • Required access: Network connectivity to the exposed communication port; no valid credentials may be required
  • Trigger point: A client connection that reaches the command-processing path
  • Why exploitation is possible: The server may accept attacker-controlled requests without authenticating or authorizing the caller, then execute supplied commands with the service account's privileges

Why this happens

The underlying issue is a missing authentication control on a network-facing command-execution service. The server binds broadly and processes requests without establishing the caller's identity or checking authorization before invoking shell commands. This behavior maps to CWE-306, Missing Authentication for Critical Function, because a security-sensitive operation is available without a required authentication step.

Potential impact

  • An unauthenticated attacker could execute arbitrary commands in the container
  • Attackers may obtain root-level control within the container
  • Bind-mounted host workspace files could be read, altered, or deleted
  • Credentials, source code, or build artifacts may be exposed
  • The container could be used for lateral movement or additional compromise

General mitigation guidance

Mitigation generally focuses on ensuring that command-execution interfaces are

authenticated, authorized, and not exposed by default.

Common defensive patterns include:

  • Requiring strong authentication and explicit authorization before processing commands
  • Binding administrative or internal services only to trusted interfaces
  • Restricting network access with segmentation, firewall policy, or service-level allowlists
  • Avoiding root execution and applying least-privilege container identities
  • Minimizing or removing host bind mounts and limiting them to read-only, narrowly scoped paths
  • Replacing shell command execution with constrained, allowlisted operations where feasible
  • Applying container isolation, resource limits, and read-only filesystem controls where practical

How teams detect issues like this

Organizations can identify this class of issue through:

  • Static analysis and code review searches for network listeners that invoke shell commands without authentication checks
  • Review of bind addresses, exposed ports, and authorization middleware in TCP server implementations
  • Container and deployment policy checks that flag services listening on all interfaces or running as root
  • Runtime network monitoring for unexpected connections to the communication port
  • Detection of unusual child processes, shell activity, or file changes in mounted workspace directories
  • Automated security review tools that enforce authentication requirements for sensitive network-facing functions

Sources

SourceLink
github.comgithub.com/HKUDS/AutoAgent
github.comgithub.com/HKUDS/AutoAgent/blob/16c12b052ef2330a198063c62a07a7f9723031e3/autoagent/environment/docker_env.py
github.comgithub.com/HKUDS/AutoAgent/blob/16c12b052ef2330a198063c62a07a7f9723031e3/autoagent/environment/tcp_server.py
github.comgithub.com/HKUDS/AutoAgent/issues/96
vulncheck.comvulncheck.com/advisories/autoagent-unauthenticated-remote-code-execution-via-the-sandbox-tcp-command-server

Frequently asked

questions

It may be remotely exploitable when the AutoAgent TCP server is reachable by an

attacker and the deployment does not provide an effective network boundary or

additional authentication layer.

The described flaw does not require application credentials because the

command-processing service may accept unauthenticated connections. Network

access to the exposed port remains a prerequisite.

Require authentication and authorization before any command-execution path,

restrict the service to trusted network interfaces, and avoid running it with

root privileges.

Treat command-execution endpoints as privileged administrative functions, apply

secure-by-default network exposure, enforce container least privilege, and

continuously test deployment policies for unintended public reachability.

Find out what's already
exploitable in your codebase.