White hat hackers just breached OpenAI using Anthropic's Claude in less than 72 hours — and it is a case study in just how fast AI is advancing

The researchers struggled to abuse the exploit with an AI agent on Opus 4.8, but the release of Opus 5 changed everything.