The vulnerability, named PwnEscape, was published on October 15. A separate researcher, 0x4d31, noted that they were able to rediscover the flaw in under 2 hours by using OpenAI's gpt-5.6-cyber model in a devbox. The rediscovery was triggered after the researcher identified the word "virtio" in a partially redacted Vercel bounty screenshot and provided that hint to the AI agent.
