Three researchers at Hacktron used Anthropic's Claude to chain an image-processing bug with a flaw in OpenAI's identity infrastructure, breaching multiple employee ChatGPT and Codex accounts in under 72 hours. One compromised Codex account even linked to OpenAI's GitHub. Wild reminder that AI isn't just a target for attacks anymore — it's becoming the tool doing the breaching, and even top labs aren't immune.