llamaCVEVulnerability Intelligence
/
Advisory · GitHub · npm · directus

GHSA-R9XQ-XP38-J4J3

Sensitive `directus_settings` fields, including plaintext AI backend authorization headers, readable by minimal `app_access` users

Parche disponible
Descargar PNG JSON
CVSS 3.1
6.5
Media
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N Fuente: GitHub
ExplotaciónSin evidenciaNo figura en CISA KEV
Exploits públicos0Ninguno conocido
Parche1 versiónPrimer parche a -22 d de la divulgación
Exposición92 dDesde la divulgación, 20 ago 2026, 11:05 a. m.
Detección90 dDesde la divulgación hasta su ingreso al índice
Timeline

Ciclo de vida de la vulnerabilidad

Hoy
Divulgación
Publicada
Explotación
Remediación
Parche 12.2.0
Monitoreo
Detectada +1
29 jul 26
20 ago 26
18 nov 26
21 nov 26
29 jul 26
27 ago 26
24 sept 26
23 oct 26
21 nov 26
FechaFaseEventoDetalle
Remediación Parche 12.2.0 directus · -22 d desde divulgación
Divulgación Publicada GHSA-R9XQ-XP38-J4J3 divulgada · CVSS 6.5 (GitHub)
Monitoreo Detectada Ingresó al índice
Monitoreo Verificada Última verificación contra las fuentes
Resumen
Descripción del advisory
Texto original · GitHub · en inglés

Summary

The built-in minimal app permissions granted unrestricted `read` access to the `directus_settings` singleton. As a
result, any authenticated user with `app_access: true` and no custom permissions could read every settings field,
including server-side AI and MCP configuration that is intended to be admin-only.

The highest impact field is `ai_openai_compatible_headers`, which stores the custom request headers used to authenticate
against a self-configured OpenAI-compatible backend. Unlike the provider API key fields, this field is not encrypted or
concealed, so any bearer token or API key an administrator placed there was returned in plaintext.

Impact

A user whose only privilege is app access could read, among others:

  • `ai_openai_compatible_headers`: plaintext authorization headers for the configured OpenAI-compatible backend
  • `ai_openai_compatible_base_url`: internal or private endpoint of that backend
  • `ai_system_prompt`: server-side AI system prompt
  • `mcp_system_prompt`: server-side MCP system prompt

The data was reachable through both the REST (`/settings`) and GraphQL (`/graphql/system`) system endpoints, since both
resolve through the same permission layer.

Where `ai_openai_compatible_headers` contained credentials, the disclosure is a usable secret rather than inert
configuration: the value was confirmed to authenticate successfully against a backend that required it. This can lead to
unauthorized use of, and billing against, the customer's AI provider account, and it discloses the internal endpoint
that backend is hosted on.

The encrypted provider key fields (`ai_openai_api_key`, `ai_openai_compatible_api_key`) were correctly masked and are
not disclosed.

Impact is reduced to disclosure of AI and MCP prompt text plus backend metadata on instances where
`ai_openai_compatible_headers` is unset or holds no credential material. Prompt disclosure itself may still reveal
operational or tenant-specific detail.

Exploitation requires a valid, non-admin account on the instance. No user interaction, no custom `directus_settings`
permission, and no additional privilege is needed.

Note for existing installations

The patch changes the built-in minimal app permissions. In versions before 12.3.0, the policy creation dialog in the
Data Studio also wrote app access permission rows into the database, so policies created that way may retain a stored
`directus_settings` read permission with no field restriction, which still exposes all fields after upgrading.
Administrators should review existing policies for an unrestricted `directus_settings` read permission and remove or
field-restrict it. Directus 12.3.0 stops the dialog from persisting these rows
(directus/directus#28101 https://github.com/directus/directus/pull/28101).

Workarounds

For instances that cannot upgrade immediately:

  • Remove credentials from `ai_openai_compatible_headers` and rotate any value that was stored there, since it must be
  • Avoid placing sensitive content in `ai_system_prompt` and `mcp_system_prompt`.

Credits

Reported by @kamil-sawicki

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

Vector CVSS 3.1

Desglose de métricas

Opción resaltada = valor del vector · izquierda = peor caso · GitHub
Vector de ataqueAV:N
RedAdyacenteLocalFísica
ComplejidadAC:L
BajaAlta
PrivilegiosPR:L
NingunoBajosAltos
InteracciónUI:N
NingunaRequerida
AlcanceS:U
CambiadoSin cambio
ConfidencialidadC:H
AltoBajoNinguno
IntegridadI:N
AltoBajoNinguno
DisponibilidadA:N
AltoBajoNinguno
Score de explotaciónQué tan fácil y probable es que la exploten
8.3/ 10Alto
Vector 8.3 (muy fácil) · sin evidencia de explotación.
Versiones

Rangos afectados

Afectada Corregida Última
directus≥ 11.15.0 < 12.2.0
11.15.0
12.2.0
12.6.1
Remediación

Estado del parche por paquete

Paquete / productoEstadoCorregida enDesde divulgaciónÚltima versión
npmdirectus
Corregida 12.2.0 -22 d 12.6.1 no afectada
Versiones corregidas12.2.0
Explotación

Arsenal público

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

Referencias
Advisoryhttps://github.com/directus/directus/security/advisories/GHSA-r9xq-xp38-j4j3

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