Analysis of AI Code Sandbox Escapes in Cohere Terrarium and OpenAI Codex CLI
CybersecurityVulnerabilitiesAISandboxEscapes
The post discusses two vulnerabilities: CVE-2026-5752 affecting Cohere Terrarium and CVE-2025-59532 affecting OpenAI Codex CLI. Both vulnerabilities involve sandbox escapes in AI code execution environments. The analysis highlights how these flaws allow malicious code to bypass security restrictions in Pyodide-based sandboxes. A linked blog post provides further technical details.