Researchers used a rival's AI model to break into OpenAI's own systems.
Security researchers used Anthropic's Claude to find and exploit vulnerabilities in OpenAI's infrastructure. The work gave them control over employee accounts and access to an internal code repository. Rather than using that access for anything malicious, the researchers reported the vulnerabilities to OpenAI. It's a disclosure, not a breach in the criminal sense, but the mechanics are the same ones an attacker would use.
The bigger story here isn't OpenAI's exposure. It's that Claude was capable enough to do the offensive security work in the first place, turning what used to require a skilled human red team into a task an AI assistant can meaningfully carry out. That should worry every company running production systems, not just the ones building chatbots.
OpenAI got the benefit of researchers who chose to disclose rather than exploit; the next company Claude, or a model like it, gets pointed at might not be so fortunate.