CYBERSECURITYApril 22, 20261 min read10 views

Cohere AI Terrarium Sandbox Flaw Enables Root Code Execution, Container Escape

A critical security vulnerability (CVE-2026-5752) with a CVSS score of 9.3 has been discovered in Cohere AI's Terrarium sandbox, potentially leading to arbitrary code execution and container escape.

A critical security vulnerability has been disclosed in a Python-based sandbox called Terrarium that could result in arbitrary code execution. The vulnerability, tracked as CVE-2026-5752, is rated 9.3 on the CVSS scoring system. "Sandbox escape vulnerability in Terrarium allows arbitrary code execution with root privileges on a host process via JavaScript prototype chain traversal," according to a description of the flaw in CVE.org. Developed by Cohere AI as an open-source project, Terrarium is a Python sandbox that's used as a Docker-deployed container for running untrusted code written by users or generated with assistance from a large language model (LLM). Notably, Terrarium runs on Pyodide, a Python distribution for the browser and Node.js, enabling it to support standard Python packages. According to the CERT Coordination Center (CERT/CC), the root cause relates to a JavaScript prototype chain traversal in the Pyodide WebAssembly environment that enables code execution with elevated privileges on the host Node.js process. Successful exploitation of the vulnerability can allow an attacker to break out of the confines of the sandbox and execute arbitrary system commands as root within the container. In addition, it can permit unauthorized access to sensitive files, such as "/etc/passwd," reach other services on the container's network, and even possibly escape the container and escalate privileges further. Security researcher Jeremy Brown has been credited with discovering and reporting the flaw. Given that the project is no longer actively maintained, the vulnerability is unlikely to be patched. As mitigations, CERT/CC is advising users to disable code submission features, segment the network, deploy a Web Application Firewall, and monitor container activity. "The sandbox fails to adequately prevent access to parent or global object prototypes, allowing sandboxed code to reference and manipulate objects in the host environment," SentinelOne said.

Source: The Hacker News

More in this category

cybersecurity

GoldenEyeDog Subgroup Linked to DigiCert Breach and Code-Signing Certificate Theft

Cybersecurity researchers have attributed the April 2026 DigiCert security incident to CylindricalCanine, a subgroup of the Chinese cybercrime group GoldenEyeDog, which compromised DigiCert's support portal to steal and abuse code-signing certificates.

cybersecurity

Armenia Detains Russian Tourist on U.S. Warrant for REvil Hacker, Lawyers Say Wrong Man

Armenia has detained a Russian tourist on a U.S. extradition warrant targeting a REvil ransomware suspect. However, defense lawyers argue it is a case of mistaken identity caused by incomplete name details in the U.S. warrant.

cybersecurity

ACR Stealer Uses ClickFix Lures to Steal Browser Tokens and Microsoft 365 Files

ACR Stealer, an active infostealer, is targeting enterprise networks using sophisticated 'ClickFix' social engineering lures. By tricking users into pasting malicious commands into the Run dialog, attackers can easily steal session tokens, browser credentials, and sensitive documents from OneDrive and SharePoint.

cybersecurity

SASE Has An AI Blind Spot. Inspecting Packets Is No Longer Enough.

Traditional SASE architectures are struggling to keep up with the shift to browser-based workflows and generative AI, as modern encryption and decentralized data interactions render packet inspection obsolete.

cybersecurity

Hackers Weaponize Balochistan Police Portal in Multi-Group Espionage Campaigns

Cybersecurity researchers have uncovered a series of cyber espionage campaigns targeting Pakistani law enforcement agencies, with suspected Chinese and Indian threat actors weaponizing a police portal to deploy malware.

cybersecurity

Study of 281 Free Android VPN Apps Finds Traffic Leaks, Unencrypted Data, and Tracking

Researchers ran 281 of the most popular free VPN apps on the Google Play Store through a new testing system and found that many fail at the basics people install a VPN for, i.e., keeping their traffic private and secure.