#sandbox
-
AI agents aren't 'going rogue' — frontier labs are just skipping security basics
Security experts who spoke with CSO Online argue that the headline-grabbing 'escapes' by AI agents aren't signs of an awakening intelligence, but the predictable result of frontier labs neglecting decades-old cybersecurity fundamentals.
-
Microsoft Ships First Release of Its LiteBox Sandboxing Library OS
Microsoft has tagged the first release of LiteBox, an open-source, Rust-based library operating system built for sandboxing applications.
-
Escaping the OpenAI Codex sandbox: two flaws that ran commands on the developer's machine
Researchers found two ways out of OpenAI Codex's sandbox — one able to run commands on a developer's machine from the strictest mode, with no prompt and nothing shown on screen. OpenAI has patched both.
-
qBittorrent “escaped its sandbox to go do things”
A tongue-in-cheek Mastodon note about a wayward torrent client racked up hundreds of votes on Hacker News, even though the post itself gives away almost nothing.
-
Auto Mode in Claude Code Is No Cure for Prompt Injection
Researcher Johann Rehberger has demonstrated an attack that defeats Claude Code's protective mode 80% of the time — and sometimes the protection itself stops the agent from shutting down malicious code.
-
Docker Unveils Sandboxes — Disposable Isolated Environments for AI Agents
Docker has announced Docker Sandboxes — disposable, isolated sandboxes designed for running AI agents. The product sparked notable discussion on Hacker News.
-
n8n Sandbox Escape: Workflow Editor Could Run OS Commands
The n8n automation platform has patched a high-severity vulnerability: an authenticated workflow editor could break out of the expression sandbox and run operating-system commands with the privileges of the n8n process.
-
The Week of Sandbox Escapes: 7 Vulnerabilities in Four AI Coding Assistants
Researchers at Pillar Security found seven ways to bypass isolation in Cursor, Codex, Gemini CLI and Antigravity. In almost every case the agent didn't have to break out of the sandbox directly — it was enough to write a file the host would later trust.