/

CVE-2026-91933

Severity:

HIGH RISK

(7.1)

Broken Access Control (CWE-639) in Flowise Realtime

CVE-2026-91933

Awaiting Analysis

Vulnerability type

Category:

Authorization Bypass

CWE:

CWE-639

Impact:

Cross-workspace unauthorized access and tool execution

Severity:

HIGH RISK

(7.1)

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

Flowise does not consistently enforce workspace-level authorization for its openai-realtime endpoints. An authenticated user who supplies an unscoped chatflow identifier may access tools belonging to another workspace, invoke them, and potentially retrieve sensitive outputs or trigger external side effects.

Affected context

The vulnerability occurs in Flowise deployments using the openai-realtime endpoints when authorization is not bound to the authenticated user's workspace. It becomes risky when an attacker can obtain or guess a ChatFlow identifier and the endpoint accepts that identifier without confirming that the referenced ChatFlow and its tools belong to the user's authorized workspace.

Exploit preconditions

  • Attack surface: The openai-realtime GET and POST endpoints
  • Required access: An authenticated Flowise account
  • Trigger point: Supplying a chatflow identifier that belongs to another workspace
  • Why exploitation is possible: The endpoint may resolve tools using an unscoped identifier without enforcing workspace ownership or membership authorization

Why this happens

The underlying bug is an object-level authorization failure. The endpoint uses a client-supplied ChatFlow identifier to retrieve or execute tools but does not reliably verify that the referenced object belongs to the authenticated user's workspace. This maps to CWE-639: Authorization Bypass Through User-Controlled Key, in which a user-controlled lookup key can expose or operate on resources outside the user's permitted scope.

Potential impact

  • Unauthorized users may retrieve tool definitions from other workspaces
  • Attackers could execute tools configured by victim workspaces
  • External systems or services may be affected by unintended tool actions
  • Sensitive tool outputs or connected data may be disclosed
  • Cross-tenant confidentiality and integrity could be compromised

General mitigation guidance

Mitigation generally focuses on enforcing authorization at the resource boundary rather than trusting identifiers supplied by clients.

Common defensive patterns include:

  • Resolve ChatFlows and tools through the authenticated user's workspace context
  • Enforce server-side ownership or membership checks for every GET and POST operation
  • Use deny-by-default authorization policies for cross-workspace resource access
  • Avoid exposing predictable or reusable resource identifiers as the sole access control mechanism
  • Apply least privilege to tools and the external services they can access
  • Record and review cross-workspace authorization failures and tool invocations

How teams detect issues like this

Organizations can look for this class of issue through:

  • Static analysis and code review checks for database lookups driven only by client-supplied identifiers
  • Review signals showing missing workspace, tenant, or ownership predicates in endpoint handlers
  • Authorization tests that attempt access to equivalent resources across workspace boundaries
  • API gateway or policy enforcement that requires tenant context for resource operations
  • Monitoring for authenticated users invoking ChatFlows or tools outside their normal workspace
  • Automated security review tools that identify insecure direct object reference patterns

Sources

SourceLink
github.comgithub.com/FlowiseAI/Flowise/security/advisories/GHSA-gggp-6qmf-xwwc
vulncheck.comvulncheck.com/advisories/flowise-before-3.1.4-authorization-bypass-via-openai-realtime

Frequently asked

questions

Yes, within the application's authenticated attack surface. An attacker generally needs network access to the affected endpoints and a valid Flowise account, but does not necessarily need access to the victim workspace.

The attacker must be authenticated and able to provide a ChatFlow identifier that resolves to a resource in another workspace. The ability to guess or obtain an identifier may affect practical exploitability.

Enforce workspace-scoped authorization on every resource lookup and tool execution, using the authenticated user's workspace and membership context rather than trusting the supplied identifier.

Use deny-by-default, object-level authorization tests for every tenant boundary, least-privilege tool permissions, and monitoring for cross-workspace access attempts.

Find out what's already
exploitable in your codebase.