criticalOther

Cua Computer-Server Unauthenticated Remote Command Execution

First seen Sep 8, 2026 · Updated Sep 8, 2026 · CVSS 9.8

agent-relevantrceauthentication-bypasscuaai-agent-infrastructurecontainer-escapeunauthenticated-access

Cua computer-server versions before 0.3.42 contain a critical authentication bypass triggered when the CONTAINER_NAME environment variable is unset, causing the service to bind to all network interfaces without requiring authentication. This exposes a TCP port 8000 service that allows unauthenticated attackers to execute arbitrary shell commands, read/write arbitrary files, and open interactive PTY shells, granting full remote control of the host or container.

Technical Analysis

The vulnerability stems from insecure default behavior in computer-server: when CONTAINER_NAME is not explicitly set, the authentication check is silently skipped and the server binds to 0.0.0.0 rather than a restricted interface. Attackers with network access to TCP port 8000 can invoke the run_command endpoint to execute arbitrary OS commands, use file operation endpoints for arbitrary file read/write, and spawn interactive PTY sessions—effectively equivalent to unauthenticated root-level shell access. The 9.8 CVSS score reflects the trivial exploitability (no authentication, no user interaction) combined with complete compromise of confidentiality, integrity, and availability. Cua's computer-server component is specifically designed as a backend for AI computer-use agents that control virtual desktops/containers, meaning organizations running autonomous AI agents via Cua are directly exposed to full takeover of the agent's execution environment, including any credentials, API keys, or tool-access tokens present on that host, making this a severe and directly agent-relevant threat.

Affected Systems

Cua computer-server versions prior to 0.3.42; deployments where the CONTAINER_NAME environment variable is unset, particularly containerized or VM-based agent execution backends exposing TCP port 8000 to untrusted networks

Indicators of Compromise

  • Unexpected inbound connections to TCP port 8000
  • Unauthenticated HTTP/API requests to run_command endpoint
  • Unauthenticated HTTP/API requests to file read/write endpoints
  • Unexpected PTY shell sessions spawned on computer-server hosts
  • Process execution originating from computer-server without corresponding authenticated session logs

Remediation Steps

  1. 1

    Upgrade computer-server

    Update Cua computer-server to version 0.3.42 or later, which enforces authentication regardless of CONTAINER_NAME configuration state.

  2. 2

    Set CONTAINER_NAME explicitly

    Ensure the CONTAINER_NAME environment variable is always explicitly set in all deployment configurations as a compensating control until upgrade is complete.

  3. 3

    Restrict network exposure

    Bind computer-server to localhost or internal-only interfaces, and use firewall rules or security groups to block external access to TCP port 8000.

  4. 4

    Audit for compromise

    Review logs for unauthenticated access to run_command, file operation, and PTY endpoints, and rotate any credentials or API keys that may have resided on affected hosts.

  5. 5

    Network segmentation for agent infrastructure

    Isolate AI agent execution backends (Cua, similar computer-use frameworks) on segmented networks with strict ingress controls to limit blast radius from similar future vulnerabilities.

CVE / Advisory IDs

CVE-2026-86121

Industries Most Exposed

TechnologyAI/ML servicesSoftware developmentCloud hostingAny industry deploying autonomous AI agent infrastructure

Sources

Respond to this threat

Pro subscribers get a full AI-generated incident-response playbook for this threat — detection, containment, eradication, and recovery steps — plus an unlimited AI Threat Advisor for questions about your environment.