Hackers Used AI to Break Into OpenAI Employee Accounts
Security researchers were able to hack into OpenAI accounts in less than 72 hours using Anthropic's Claude AI. They gained access to sensitive internal repositories, raising serious questions about AI security.
A team of three independent security researchers at a company called Hacktron discovered something alarming: they could break into OpenAI employee accounts in less than three days using Claude Opus (an advanced AI model made by Anthropic, a competitor to OpenAI).
They were able to access OpenAI's GitHub repository — think of it as a digital vault where companies store their most important code and technical secrets. The researchers didn't actually download the secret code themselves, but they proved they could get in by sending a test request from a stolen employee account.
What makes this story important: it shows that even the world's most advanced AI companies have security vulnerabilities. The fact that one AI company's tool could be used to hack another AI company highlights a real-world problem. As AI becomes more powerful, the stakes get higher — not just for tech companies, but for anyone who trusts these services with their data.
The researchers responsibly disclosed their findings rather than exploiting them for harm, which is how security research should work. But it's a wake-up call that AI security needs constant attention.
Original source: The Verge AI