Security researchers used Claude to help them hack into OpenAI
Executive Take
AI tools now let small teams break into major AI labs faster than internal security can react. Every company using third-party services like Discourse needs to treat them as part of its core attack surface.
Executive Summary
Three independent researchers at Hacktron used Anthropic's Claude Opus models to hack into OpenAI employee accounts within 72 hours, per The Wall Street Journal. They accessed OpenAI's Discourse-hosted community forum and from there reached a Codex account, sending a pull request to prove they could reach OpenAI's "Monorepo" GitHub repository without opening its code.
Why It Matters
Technology and cybersecurity leaders should note that AI is now a hacking accelerant, not just a target. A three-person team breached a top AI lab's internal systems in under three days using a rival's AI model.
Bizquad Perspective
The real story isn't that OpenAI got hacked, it's that the entry point was a vendor's forum software, proof that most breaches now start outside the company's own walls.