> TritonInfoSec News
Home / Sep 21, 2026 / Story
0
#3 BleepingComputer general September 20, 2026 at 12:00 UTC

Researchers escape OpenAI Codex sandbox to run commands on host

By Ax Sharma

AI Summary

Researchers discovered two sandbox escape vulnerabilities in OpenAI's Codex, one of which allowed execution of arbitrary commands on a developer's host machine even from Codex's most restrictive isolation mode. OpenAI has patched both vulnerabilities, but the findings expose critical risks in AI coding assistant deployments where sandbox integrity is assumed.

Read full article → https://www.bleepingcomputer.com/news/security/res…

Relevance score: 83.0/100

# More from September 21

  1. 1
    An undercover Google analyst infiltrated a notorious supply-chain hacking gang Ars Technica Security
  2. 2
    Google Gemini Broke Into Real Company Systems After Security Test Domain Mix-Up The Hacker News
  3. 4
    BragJack attacks hijack AI browser agents through malicious extensions BleepingComputer
  4. 5
    Malicious npm packages evade install-script defenses at runtime BleepingComputer
  5. 6
    Calling viral AI actress Tilly Norwood? Agree to a face scan first BleepingComputer
  6. 7
    Identity Visibility in 2026: The Foundation of Identity Security The Hacker News
  7. 8
    TigerByte Cyber Emerges From Stealth With $3 Million in Funding SecurityWeek
  8. 9
    ‘Nudify’ apps: What to do if someone makes a fake nude of you WeLiveSecurity (ESET)
  9. 10
    Viral AI actress' hotline face-scans every caller, watches their mood BleepingComputer