CVE-2026-54769

Langroid: Sandbox Escape to Remote Code Execution via Incomplete `eval()` Mitigation in TableChatAgent

Publicada 10 jul 2026Modificada 10 jul 2026CWE 94

En la Trivy database desde, como mínimo, el 24 oct 2026, 07:11 UTCNo consta cuándo entró antes de ese build.

Severidad por fuente

GHSA10.0CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H

La etiqueta propia de Trivy es CRITICAL. Crítica según GHSA.

Descripción

Langroid is a framework for building large-language-model-powered applications. Versions prior to 0.65.2 are vulnerable to a critical Sandbox Escape leading to Remote Code Execution (RCE) in its `TableChatAgent` and `VectorStore` capabilities. When these agents evaluate LLM-generated tool messages with `full_eval=True`, they attempt to sandbox the execution by explicitly setting `locals` to an empty dictionary `{}` inside Python's `eval()` function. However, this relies on an incomplete understanding of Python's execution model. Because `__builtins__` is not explicitly scrubbed from the `globals` dictionary mapping, Python implicitly injects all built-ins during execution, granting full access to functions like `__import__('os').system()`. Since `TableChatAgent.pandas_eval()` executes external LLM outputs natively, this bypass permits any attacker providing prompt payload to achieve unauthenticated RCE on the host system. Version 0.65.2 patches the issue.

Paquetes afectados y versiones corregidas

Fuente	Paquete	Corregido en
pip::GitHub Security Advisory pip	langroid	0.65.2

Referencias

Qué hace Attestkeep con esto

Dentro de su cluster, sin abrir una hoja de cálculo.

SabeEl scanner actualiza su vulnerability database cada seis horas.
EncuentraCada image admitida se vuelve a escanear cada 24 horas por defecto (scan.rescanInterval). Cuando Trivy califica esta vulnerabilidad como crítica para la fuente de paquetes de la image, aparece por su nombre en los findings de la image. Los findings sin corrección quedan fuera salvo que includeUnfixed esté activado.
AvisaUn evento critical_finding va a los canales suscritos a él: e-mail, Slack, Teams, un webhook, Jira, GitLab o GitHub.
RechazaCon una policy que bloquea los findings críticos, un pod con esta image se rechaza en la admission, incluido un rollback a ella. Los pods que ya están en marcha no se desalojan.
RegistraCada decisión se guarda con la policy que se aplicó y su hash, y se sella bajo un checkpoint firmado.

Probar Attestkeep Leer la documentación

Vulnerability data compiled from the Trivy vulnerability database (trivy-db, Aqua Security, Apache License 2.0), which aggregates third-party advisories. CVE records are published by The MITRE Corporation under the CVE Terms of Use; CVE is a trademark of MITRE. This product uses data from the NVD API but is not endorsed or certified by the NVD. Severity, titles and advisory data from the GitHub Advisory Database (CC BY 4.0), Red Hat security data (CC BY 4.0) and the other sources linked on each entry. Entries may be summarised or re-formatted; the linked original is the authoritative record.