Skip to main content
Vulnerability Database/CVE-2026-111321

CVE-2026-111321: TaskingAI SSRF Vulnerability

CVE-2026-111321 is a server-side request forgery vulnerability in TaskingAI through version 0.3.0 that enables authenticated attackers to force the backend to send arbitrary HTTP requests. This article covers technical details, affected versions, security impact, and mitigation strategies.

Published:

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-2026-111321 Overview

CVE-2026-111321 is a Server-Side Request Forgery (SSRF) vulnerability [CWE-918] affecting TaskingAI through version 0.3.0. The flaw lets authenticated attackers coerce the TaskingAI backend into issuing arbitrary HTTP requests. The root cause is an unvalidated servers[0].url field in user-defined OpenAPI action definitions.

Attackers holding a project API key or console admin login can create an action pointing to loopback, internal, or cloud metadata addresses. Triggering the action through POST /actions/{action_id}/run returns up to 64 KiB of the response body to the caller. This enables reconnaissance of internal services and potential retrieval of cloud instance metadata.

Critical Impact

Authenticated attackers can pivot through the TaskingAI backend to reach internal network services and cloud metadata endpoints, exfiltrating up to 64 KiB per request.

Affected Products

  • TaskingAI through version 0.3.0
  • Backend component openapi_call.py in the action execution service
  • Backend component openapi_utils.py handling OpenAPI server URL parsing

Discovery Timeline

  • 2026-10-15 - CVE-2026-111321 published to NVD
  • 2026-10-16 - Last updated in NVD database

Technical Details for CVE-2026-111321

Vulnerability Analysis

TaskingAI exposes a feature allowing users to define actions from OpenAPI specifications. The specification includes a servers array whose first element supplies the base URL for HTTP calls originating from the backend. The backend accepts this URL without validating the host, scheme, or address range.

When a client issues POST /actions/{action_id}/run, the service in openapi_call.py constructs an outbound HTTP request to the attacker-controlled target. The handler returns the response body back to the caller, truncated to 64 KiB. There is no allowlist, no block on private address space, and no interception of DNS rebinding or redirects to internal addresses.

An authenticated attacker with either a project API key or console admin credentials can abuse this trust boundary. The vulnerability turns the backend into an unauthenticated proxy into any network segment reachable by the TaskingAI deployment.

Root Cause

The root cause is missing input validation on the servers[0].url field during OpenAPI action creation and execution. The relevant code paths in backend/app/services/tool/action/openapi_call.py (line 144) and backend/app/services/tool/action/openapi_utils.py (lines 168-169) parse and dispatch the URL without filtering internal, loopback, link-local, or cloud metadata ranges such as 169.254.169.254.

Attack Vector

An authenticated attacker performs the following steps. First, the attacker authenticates with a valid project API key or console admin session. Second, the attacker creates a user-defined action whose OpenAPI spec sets servers[0].url to a target such as http://127.0.0.1:8080, http://169.254.169.254/latest/meta-data/, or an internal RFC1918 host. Third, the attacker calls POST /actions/{action_id}/run and reads back up to 64 KiB of the HTTP response. See the VulnCheck SSRF Advisory for TaskingAI and the HackMD SSRF Guide for TaskingAI for additional technical detail.

Detection Methods for CVE-2026-111321

Indicators of Compromise

  • Action definitions whose OpenAPI servers[0].url points to loopback (127.0.0.0/8), link-local (169.254.0.0/16), or RFC1918 ranges.
  • Outbound HTTP connections from the TaskingAI backend process to cloud metadata endpoints such as 169.254.169.254 or metadata.google.internal.
  • Repeated calls to POST /actions/{action_id}/run for actions created shortly beforehand by the same API key.

Detection Strategies

  • Audit stored OpenAPI action definitions and flag any servers[0].url value whose resolved IP falls into internal or metadata ranges.
  • Correlate action creation events with immediate execution events from the same identity as a probable SSRF probe pattern.
  • Monitor backend egress with a proxy or sidecar and alert on requests to addresses outside the published allowlist.

Monitoring Recommendations

  • Log all POST /actions/{action_id}/run calls with associated action metadata including the destination host and response size.
  • Capture DNS resolutions performed by the TaskingAI backend to detect DNS rebinding attempts resolving public names to internal IPs.
  • Alert on anomalous spikes in action execution volume from a single project API key.

How to Mitigate CVE-2026-111321

Immediate Actions Required

  • Restrict issuance of project API keys and console admin access to trusted operators only, since the vulnerability requires authentication.
  • Place the TaskingAI backend behind an egress proxy that denies connections to loopback, link-local, RFC1918, and cloud metadata ranges.
  • Review existing user-defined actions and remove any whose servers[0].url targets internal addresses.

Patch Information

At time of publication, no fixed version is listed in the NVD entry. The vulnerability is confirmed in TaskingAI through 0.3.0. Monitor the GitHub TaskingAI Repository for updated releases addressing the SSRF in openapi_call.py and openapi_utils.py.

Workarounds

  • Deploy the TaskingAI backend in a network namespace that cannot reach internal services or the cloud metadata endpoint.
  • Enforce Instance Metadata Service v2 (IMDSv2) on AWS workloads to require session tokens that the SSRF request flow cannot obtain.
  • Apply a reverse proxy or WAF rule rejecting action payloads whose servers values resolve to disallowed address ranges.
bash
# Example iptables egress policy for the TaskingAI backend container
iptables -A OUTPUT -d 127.0.0.0/8      -j REJECT
iptables -A OUTPUT -d 169.254.0.0/16   -j REJECT
iptables -A OUTPUT -d 10.0.0.0/8       -j REJECT
iptables -A OUTPUT -d 172.16.0.0/12    -j REJECT
iptables -A OUTPUT -d 192.168.0.0/16   -j REJECT

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

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.