/

CVE-2026-51903

Severity:

HIGH RISK

(8.1)

Broken Access Control (CWE-284) in SuperAGI Agent API

CVE-2026-51903

Deferred

Vulnerability type

Category:

Broken Access Control

CWE:

CWE-284

Impact:

Cross-organization information disclosure

Severity:

HIGH RISK

(8.1)

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

SuperAGI's agent-execution endpoint may return the latest execution and agent metadata for a project belonging to another organization. An authenticated attacker may access this information by supplying a victim project ID when authorization is not properly scoped to the caller's organization.

SuperAGI's agent-execution endpoint may return the latest execution and agent metadata for a project belonging to another organization. An authenticated attacker may access this information by supplying a victim project ID when authorization is not properly scoped to the caller's organization.

Affected context

The issue is described in get_agent_by_latest_execution in superagi/controllers/agent_execution.py, exposed through /agentexecutions/get/latest/agent/project/{project_id}. It may affect deployments where the endpoint accepts a project ID without verifying that the authenticated user's organization is authorized to access that project.

The issue is described in get_agent_by_latest_execution in superagi/controllers/agent_execution.py, exposed through /agentexecutions/get/latest/agent/project/{project_id}. It may affect deployments where the endpoint accepts a project ID without verifying that the authenticated user's organization is authorized to access that project.

Exploit preconditions

  • Attack surface: The remotely accessible latest-agent-execution endpoint
  • Required access: An authenticated account in an organization
  • Trigger point: A request to the endpoint using a project ID associated with another organization
  • Why exploitation is possible: The project identifier may be used to retrieve execution and agent metadata without an effective organization-level authorization check
  • Attack surface: The remotely accessible latest-agent-execution endpoint
  • Required access: An authenticated account in an organization
  • Trigger point: A request to the endpoint using a project ID associated with another organization
  • Why exploitation is possible: The project identifier may be used to retrieve execution and agent metadata without an effective organization-level authorization check

Why this happens

The underlying issue is improper access control: authorization may not be enforced against the requested project's organization before returning its data. This maps to CWE-284, which covers failures to restrict access to resources according to the caller's permissions. Treating possession of a project ID as sufficient authority is an insecure access-control pattern.

The underlying issue is improper access control: authorization may not be enforced against the requested project's organization before returning its data. This maps to CWE-284, which covers failures to restrict access to resources according to the caller's permissions. Treating possession of a project ID as sufficient authority is an insecure access-control pattern.

Potential impact

  • An authenticated user may be able to view another organization's latest execution metadata
  • Agent or project details could be exposed across organizational boundaries
  • Disclosure may support further targeting of affected projects or organizations
  • An authenticated user may be able to view another organization's latest execution metadata
  • Agent or project details could be exposed across organizational boundaries
  • Disclosure may support further targeting of affected projects or organizations

General mitigation guidance

Defensive patterns include enforcing authorization on every request and scoping resource lookups to the authenticated user's organization. Treat project IDs as identifiers, not proof of permission, and return only data the caller is authorized to access. Prefer deny-by-default authorization, consistent tenant isolation, and least-privilege access to execution metadata.

Defensive patterns include enforcing authorization on every request and scoping resource lookups to the authenticated user's organization. Treat project IDs as identifiers, not proof of permission, and return only data the caller is authorized to access. Prefer deny-by-default authorization, consistent tenant isolation, and least-privilege access to execution metadata.

How teams detect issues like this

  • Review endpoint handlers and data-access paths for authorization checks performed before returning project or execution data.
  • Use static analysis or code review to identify resource lookups based on caller-supplied identifiers without tenant scoping.
  • Test authorization policies with accounts from separate organizations and verify cross-tenant requests are denied.
  • Monitor for requests where a user accesses project IDs outside their organization, subject to privacy and logging policies.
  • Review endpoint handlers and data-access paths for authorization checks performed before returning project or execution data.
  • Use static analysis or code review to identify resource lookups based on caller-supplied identifiers without tenant scoping.
  • Test authorization policies with accounts from separate organizations and verify cross-tenant requests are denied.
  • Monitor for requests where a user accesses project IDs outside their organization, subject to privacy and logging policies.

Sources

SourceLink
gist.github.comgist.github.com/Ro1ME/d5e2ff57f9ed891bac0d401c0a7e9aae
github.comgithub.com/TransformerOptimus/SuperAGI

Frequently asked

questions

The described endpoint is remotely accessible, but exploitation requires authentication and a project ID belonging to another organization.

Yes. The description specifies a remote authenticated attacker; the account's organization is distinct from the victim project's organization.

Verify that the authenticated caller is authorized for the requested project before retrieving or returning its execution and agent metadata.

Apply deny-by-default, tenant-aware authorization consistently across endpoints, and test access boundaries using identities from different organizations.

The described endpoint is remotely accessible, but exploitation requires authentication and a project ID belonging to another organization.

Yes. The description specifies a remote authenticated attacker; the account's organization is distinct from the victim project's organization.

Verify that the authenticated caller is authorized for the requested project before retrieving or returning its execution and agent metadata.

Apply deny-by-default, tenant-aware authorization consistently across endpoints, and test access boundaries using identities from different organizations.

Find out what's already
exploitable in your codebase.