AIBleepingComputerabout 12 hours ago
AI coding agent sandboxes bypassed in four tools
1 min read
Security researchers broke out of sandboxes in Cursor, OpenAI's Codex, Google's Gemini CLI, and Antigravity. The agent writes a file that a trusted tool outside the sandbox later runs. Prompt injection triggers the escape. Cursor's bug is CVE-2026-48124, fixed in version 3.0.0. Codex CLI was patched in v0.95.0.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyGoogle
Summary by ByteBrief
