/

CVE-2026-72922

Severity:

HIGH RISK

(8.2)

Authentication Bypass (CWE-287) in AutoGPT Webhooks

CVE-2026-72922

Deferred

Vulnerability type

Category:

Authentication Bypass

CWE:

CWE-287

Impact:

Unauthorized workflow execution and potential owner-context actions

Severity:

HIGH RISK

(8.2)

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

AutoGPT contains an authentication bypass in its generic webhook ingress handling, where the selected manager may not match the webhook's configured provider. When a configured secret token is present, an attacker may bypass signature verification and execute the associated generic webhook graph as its owner.

Affected context

The issue occurs in AutoGPT deployments using the webhook ingress route for Compass webhook identifiers. Risk exists when the provider value is supplied through the request URL, the webhook is configured with a secret token, and manager selection is not bound to the webhook's stored provider metadata.

Exploit preconditions

  • Attack surface: AutoGPT's webhook ingress HTTP endpoint
  • Required access: Ability to send requests to a known or reachable webhook ingress URL
  • Trigger point: Processing of a generic webhook request for a Compass-associated webhook identifier
  • Why exploitation is possible: The provider selected from the URL can cause a manager with a no-op signature verification method to handle the request, bypassing the configured secret-token check

Why this happens

The vulnerable route derives webhook-manager selection from an untrusted URL segment without verifying that it matches the webhook's configured provider. This permits a manager whose inherited verification method performs no effective signature check to process a request that should require authentication. The behavior maps to CWE-287, Improper Authentication, because the system fails to reliably verify that an incoming webhook request is authorized before executing owner-associated workflow logic.

Potential impact

  • An attacker may invoke a generic webhook graph without the configured secret token.
  • Unauthorized workflows could execute with the webhook owner's effective context.
  • Connected services or data may be accessed or modified according to workflow permissions.
  • The affected automation environment could be used for further unauthorized actions.

General mitigation guidance

Mitigation generally focuses on binding authentication decisions to trusted

webhook configuration rather than request-controlled routing values.

Common defensive patterns include:

  • Selecting the webhook manager from server-side webhook metadata.
  • Requiring the resolved provider and requested provider to match before processing.
  • Enforcing signature verification for every authenticated webhook path.
  • Treating no-op verification implementations as unsuitable for protected endpoints.
  • Applying least-privilege permissions to webhook-owned workflows and integrations.
  • Separating unauthenticated ingestion from privileged workflow execution.

How teams detect issues like this

Organizations can look for this class of issue through:

  • Static analysis and code review checks for security-sensitive dispatch based on URL parameters.
  • Review signals where authentication or signature verification occurs after provider or handler selection.
  • Tests that submit mismatched provider identifiers and confirm requests are rejected.
  • Policy enforcement requiring server-side identity binding for webhook routing.
  • Monitoring for webhook executions lacking expected signature-validation events.

Automated review tools may help identify inconsistent authentication paths and

no-op security implementations.

Sources

SourceLink
github.comgithub.com/Significant-Gravitas/AutoGPT/commit/646dd5b8cfad1206e92ec7bcc3b8312657e2a92e
github.comgithub.com/Significant-Gravitas/AutoGPT/pull/13559
github.comgithub.com/Significant-Gravitas/AutoGPT/releases/tag/autogpt-platform-beta-v0.6.70
github.comgithub.com/Significant-Gravitas/AutoGPT/security/advisories/GHSA-349p-3c3r-8mjr

Frequently asked

questions

It may be remotely exploitable when the webhook ingress endpoint is reachable

by an attacker and the attacker can identify or obtain a relevant webhook

identifier.

The described bypass is intended to defeat the configured webhook secret

check, so prior authentication may not be required. Network reachability and

a suitable webhook identifier remain important conditions.

Resolve the provider from trusted server-side webhook metadata, require an

exact provider match, and perform mandatory signature verification before

invoking workflow logic.

Test every authentication path, prohibit request-controlled security

handler selection, monitor rejected and accepted webhook signatures, and

apply least privilege to workflow integrations.

Find out what's already
exploitable in your codebase.