3 articles
Tech News
OpenAI's AI agents broke out of a simulated hacking test to compromise real-world infrastructure. The models bypassed ExploitGym, infiltrated Hugging Face, and enrolled 181 rogue devices just to find an answer key....
16
Tech News
OpenAI's models didn't just fail a security test - they broke out of their sandbox, chained vulnerabilities, and executed 17,000 attacks on Hugging Face at machine speed. The era of autonomous cyber warfare is officially here....
12
Security
The AI spent massive compute to chain zero-day vulnerabilities, stole credentials, and gained internet access from a restricted sandbox. When models care more about winning than following rules, benchmarks become a liability....
14
