Security researchers used Claude to help them hack into OpenAI
News Source
•2026-09-18T11:30:16-04:00
📰 What Happened
A team of three security researchers at Hacktron say they hacked into OpenAI accounts. The Wall Street Journal reported the story. They used Anthropic's Claude Opus 4.8 and 5 to help them. It took less than 72 hours.
The team reached OpenAI's private GitHub repository, known as "Monorepo". It is said to hold the firm's algorithmic secrets. They got in through Discourse, the service that hosts OpenAI's community forums. They abused how it handled HEIF image files. They stopped short of reading the secret code.
🔍 The Backstory
AI models are now very good at finding and using software flaws. According to Hacktron, Claude Opus 5 launched on the evening of July 24. By 10am the next day, the team had used it to break into the forum system. They sent a pull request from an employee's Codex account to prove access.
The researchers were testing how easy AI makes hacking. Their work shows how fast an AI helper can turn small flaws into big breaks. OpenAI is one of the most closely watched tech firms in the world. Its own tools were built to be safe, yet its systems were still breached.
🎯 Why It Matters
Most of us use online accounts run by big tech firms. This shows how AI can speed up attacks and why companies must patch software fast.
A team of three security researchers at Hacktron say they hacked into OpenAI accounts. The Wall Street Journal reported the story. They used Anthropic's Claude Opus 4.8 and 5 to help them. It took less than 72 hours.
The team reached OpenAI's private GitHub repository, known as "Monorepo". It is said to hold the firm's algorithmic secrets. They got in through Discourse, the service that hosts OpenAI's community forums. They abused how it handled HEIF image files. They stopped short of reading the secret code.
AI models are now very good at finding and using software flaws. According to Hacktron, Claude Opus 5 launched on the evening of July 24. By 10am the next day, the team had used it to break into the forum system. They sent a pull request from an employee's Codex account to prove access.
The researchers were testing how easy AI makes hacking. Their work shows how fast an AI helper can turn small flaws into big breaks. OpenAI is one of the most closely watched tech firms in the world. Its own tools were built to be safe, yet its systems were still breached.
Most of us use online accounts run by big tech firms. This shows how AI can speed up attacks and why companies must patch software fast.