/

CVE-2026-48766

Severity:

HIGH RISK

(7.6)

API Key Exposure (CWE-200) in TypeBot OpenAI Helper

CVE-2026-48766

Deferred

Vulnerability type

Category:

Sensitive Information Exposure

CWE:

CWE-200

Impact:

Exfiltration of stored API credentials

Severity:

HIGH RISK

(7.6)

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

TypeBot versions before 3.17.0 contain an information exposure vulnerability in the OpenAI model-listing helper. A low-privilege guest workspace member may supply an attacker-controlled endpoint and cause the server to send a stored OpenAI-compatible API key to that endpoint.

Affected context

The issue occurs in TypeBot workspaces where guests can read workspace membership data and invoke the OpenAI model-listing functionality. The vulnerable flow decrypts a selected workspace credential, places the secret in the OpenAI client configuration and request headers, and uses a caller-controlled baseUrl for the outbound request.

Exploit preconditions

  • Attack surface: The OpenAI model-listing helper and its outbound HTTP request path
  • Required access: Low-privilege guest membership in the affected workspace
  • Trigger point: A model-listing request that supplies an attacker-controlled destination URL and references an accessible credential
  • Why exploitation is possible: Permission checks allow readable workspace members to reach the helper, while credential identifiers are exposed and the server forwards the decrypted secret to the supplied endpoint

Why this happens

The vulnerability results from combining insufficient authorization with attacker-controlled outbound request routing. The application treats workspace readability as sufficient authorization to use a credential and then includes the decrypted secret in a request sent to a caller-supplied URL. This behavior maps to CWE-200: Exposure of Sensitive Information to an Unauthorized Actor, because a protected API key can be disclosed outside its intended trust boundary.

Potential impact

  • An attacker may obtain stored OpenAI-compatible API keys
  • Exposed credentials could allow unauthorized use of external AI services
  • Organizations may incur unexpected provider charges
  • Attackers could use the keys to access associated models or provider resources
  • Credential exposure could enable further compromise of connected systems

General mitigation guidance

Mitigation generally focuses on enforcing credential-specific authorization and preventing secrets from being sent to destinations controlled by untrusted callers.

Common defensive patterns include:

  • Restricting credential use to explicitly authorized roles and workspace capabilities
  • Avoiding caller-controlled base URLs for requests that contain secrets
  • Allowing outbound requests only to approved provider domains or configured endpoints
  • Keeping credential identifiers and secret-bearing operations inaccessible to guest members
  • Redacting API keys from logs, responses, errors, and client-visible metadata
  • Applying least privilege, short-lived credentials, and provider-side key scoping where supported
  • Rotating potentially exposed credentials and reviewing their usage through provider audit records

How teams detect issues like this

Organizations can look for this class of issue through:

  • Static analysis rules that flag secrets placed in requests whose destinations derive from user input
  • Code review checks for authorization decisions based only on workspace visibility or membership
  • Tests verifying that guest users cannot enumerate or invoke secret-bearing credential operations
  • Egress policies restricting service requests to approved provider domains
  • Monitoring for unusual API-key usage, unexpected destination hosts, and anomalous model-listing traffic
  • Automated security review tools that identify credential flows crossing untrusted trust boundaries

Sources

SourceLink
github.comgithub.com/baptisteArno/typebot.io/commit/7ae4c007d0987d2ca907b47e1b7418db62b8a157
github.comgithub.com/baptisteArno/typebot.io/pull/2459
github.comgithub.com/baptisteArno/typebot.io/releases/tag/v3.17.0
github.comgithub.com/baptisteArno/typebot.io/security/advisories/GHSA-gc3v-9whw-6wjh

Frequently asked

questions

It may be exploitable through the TypeBot application interface by a guest workspace member, provided the vulnerable helper and a readable credential are available.

The described path requires low-privilege guest membership in a workspace. The attacker also needs the ability to invoke the model-listing helper and influence its destination URL.

Enforce authorization at the individual credential and operation level, and never send decrypted secrets to caller-controlled destinations. Outbound destinations should be fixed or restricted to an approved allowlist.

Treat API keys as high-sensitivity resources, minimize their exposure to guest roles, validate outbound destinations, apply egress controls, and test every credential-bearing endpoint across privilege levels.

Find out what's already
exploitable in your codebase.