OpenAI Models Escaped and Hacked a Company in Cybersecurity Test Gone Wrong

W wsj.com ↗

▲ 29 points • 4 comments • by flippyhead • 4w ago • HN discussion ↗

Pangram verdict · v3.3

We believe that this document is fully human-written

0 %

AI likelihood · overall

Human

100% human-written 0% AI-generated

SEGMENTS · HUMAN 1 of 1

SEGMENTS · AI 0 of 1

WORD COUNT 70

PEAK AI % 0% · §1

Analyzed

Jul 22

backend: pangram/v3.3

Segments scanned

1 windows

avg 70 words each

Distribution

100 / 0%

human / AI fraction

Verdict

Human

Pangram v3.3

Article text · 70 words · 1 segments analyzed

Human AI-generated

§1 Human · 0%

By Robert McMillan and Amrith RamkumarOpenAI had caged the test models in a ‘sandbox,’ a system that didn’t have access to the internet. SeongJoon Cho/Bloomberg NewsIt’s the stuff of cybersecurity nightmares. On Tuesday, OpenAI said two artificial intelligence systems it was testing broke out of their test environment, hacked their way onto the internet and broke into another company. Copyright ©2026 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8