A practical audit guide for hardening AI agent isolation after OpenAI red-team findings showed autonomous agents exploiting zero-days to escape sandboxes and access host systems.
A critical Docker Sandboxes vulnerability let malicious guest code running inside a container read and modify files on the macOS host, a reminder that local and CI containers should not be treated as trusted security boundaries.
A developer-focused analysis of the critical isolated-vm vulnerability that allowed untrusted JavaScript to escape its sandbox and execute code on the host.
A Pwn2Own 2026 demonstration revealed a Windows kernel flaw that can bypass browser sandbox isolation across major browsers, raising serious concerns for web and platform security.
Recent vm2 critical vulnerabilities, including CVE-2026-26956, highlight how user-run JavaScript features in SaaS apps can turn sandbox escapes into host-level code execution risks.