Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-101065 Overview
CVE-2026-101065 is a critical missing authentication vulnerability [CWE-306] in Obot, an open-source AI agent and Model Context Protocol (MCP) platform. The documented Docker quickstart command binds the container to 0.0.0.0:8080 with authentication disabled by default. Every incoming request is mapped to a synthetic nobody user that holds Owner and Admin roles. Any unauthenticated party reaching the exposed port gains full administrative control over the Obot API and UI. Because the quickstart mounts /var/run/docker.sock into the container, attackers can also reach the host's Docker control surface through attacker-controlled MCP servers.
Critical Impact
Unauthenticated remote attackers obtain full administrative access to Obot and can pivot to host-level Docker control via the mounted socket.
Affected Products
- Obot AI agent/MCP platform — all versions up to and including commit d7e6970
- Deployments launched via the README Docker quickstart command
- Instances exposing port 8080 to untrusted networks without OBOT_SERVER_ENABLE_AUTHENTICATION=true
Discovery Timeline
- 2026-09-27 - CVE-2026-101065 published to NVD
- 2026-09-30 - Last updated in NVD database
Technical Details for CVE-2026-101065
Vulnerability Analysis
The Obot server supports an unauthenticated mode intended for local development. When authentication is disabled, the request pipeline substitutes a synthetic nobody principal for every caller. That principal is granted the Owner and Admin roles, bypassing all authorization checks on the API and web console.
The documented Docker quickstart deploys the server with this mode enabled while binding the listener to 0.0.0.0:8080. Any network-reachable client can register MCP servers, launch agents, and modify platform state without presenting credentials. The quickstart additionally bind-mounts /var/run/docker.sock into the container so the MCP runtime can spawn tool backends. An attacker who registers a malicious MCP server can therefore issue arbitrary Docker commands against the host, including launching privileged containers to escape to the host filesystem.
Root Cause
The root cause is an insecure default configuration combined with a documentation defect. Authentication is off by default in the shipped quickstart, and the synthetic fallback user holds administrative roles rather than minimal privilege. The fix is documentation-only: the updated quickstart sets OBOT_SERVER_ENABLE_AUTHENTICATION=true before exposing the port.
Attack Vector
Exploitation requires only network reachability to the exposed Obot port. The attacker sends unauthenticated HTTP requests to the administrative API, registers an attacker-controlled MCP server, and triggers the MCP runtime to execute commands through the mounted Docker socket. No user interaction, credentials, or prior access are required.
See the GitHub Security Advisory GHSA-jj4w-pfgv-4mrm and the VulnCheck Advisory on Unauthenticated Access for additional technical context.
Detect Identity Attacks the Moment Authentication Fails
Singularity™ Identity surfaces credential misuse, lateral movement, and privilege escalation in real time, so you can contain the breach before it spreads.
Stop Identity AttacksDetection Methods for CVE-2026-101065
Indicators of Compromise
- Unauthenticated HTTP requests to /api/* endpoints on Obot port 8080 from external addresses
- Newly registered MCP server entries pointing to attacker-controlled hostnames or IPs
- Unexpected container creation events on the Docker host, especially containers run with --privileged or bind-mounting host paths
- Audit log entries attributing administrative actions to the synthetic nobody user
Detection Strategies
- Inventory Obot deployments and verify the value of the OBOT_SERVER_ENABLE_AUTHENTICATION environment variable at runtime
- Query network telemetry for inbound connections to port 8080 on hosts running the Obot container image
- Correlate Obot application logs with Docker daemon logs to flag MCP-driven container launches that follow unauthenticated API calls
Monitoring Recommendations
- Alert on any Obot container running with /var/run/docker.sock mounted and listening on a non-loopback interface
- Monitor Docker API events (container.create, exec_create) originating from MCP runtime processes
- Track creation of new users, agents, or MCP server registrations within short windows after deployment
How to Mitigate CVE-2026-101065
Immediate Actions Required
- Set OBOT_SERVER_ENABLE_AUTHENTICATION=true on every Obot instance before exposing it to any untrusted network
- Restrict the listener to 127.0.0.1:8080 or place the service behind an authenticated reverse proxy if public exposure is not required
- Audit MCP server registrations and remove any entries that cannot be attributed to a known operator
- Rotate any secrets, API tokens, or model credentials that were accessible to the Obot instance
Patch Information
No code patch is required. The maintainers issued a documentation-only fix updating the README quickstart to enable authentication by default. Operators who followed previous instructions must manually set OBOT_SERVER_ENABLE_AUTHENTICATION=true and redeploy. Reference the GitHub Security Advisory GHSA-jj4w-pfgv-4mrm for the updated guidance.
Workarounds
- Bind the Obot container to the loopback interface only and tunnel access through SSH or a VPN
- Remove the /var/run/docker.sock bind mount when the MCP runtime does not require host Docker access
- Place Obot behind a reverse proxy that enforces mutual TLS or SSO before requests reach the application
# Configuration example: enable authentication and avoid exposing Docker socket
docker run -d \
-e OBOT_SERVER_ENABLE_AUTHENTICATION=true \
-p 127.0.0.1:8080:8080 \
--name obot \
ghcr.io/obot-platform/obot:latest
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

