Sandbox Escape Vulnerability in Langroid Framework by Langroid
CVE-2026-54769

10CRITICAL

Key Information:

Vendor

Langroid

Status
Langroid
Vendor
CVE Published:
9 July 2026

What is CVE-2026-54769?

The Langroid framework, designed for developing large-language-model-powered applications, has a vulnerability that results from an inadequate sandboxing process in its TableChatAgent and VectorStore functionalities. Versions earlier than 0.65.2 are susceptible to this issue, where LLM-generated tool messages evaluated with full_eval=True expose the system to Remote Code Execution (RCE). The framework attempts to secure execution by setting locals to an empty dictionary within Python's eval() function; however, this approach fails to properly manage the execution context. Specifically, the __builtins__ remain accessible, which allows unauthorized RCE commands through standard Python imports. Consequently, an attacker can execute arbitrary code on the vulnerable host system if they craft a suitable input prompt. This flaw has been addressed in version 0.65.2.

Affected Version(s)

langroid < 0.65.2

References

https://github.com/langroid/langroid/security/advisories/...
x_refsource_CONFIRM

CVSS V3.1

Score:
10
Severity:
CRITICAL
Confidentiality:
High
Integrity:
High
Availability:
High
Attack Vector:
Network
Attack Complexity:
Low
Privileges Required:
None
User Interaction:
None
Scope:
Changed

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.