A Discord Group Accessed Anthropic’s Most Dangerous AI

In this AI news episode, we break down the Anthropic Mythos leak, where a Discord group allegedly accessed the powerful cybersecurity model by guessing a URL, and what that says about AI security, model leaks, and frontier AI governance. We also cover a new AI medicine study showing large language models struggling with diagnosis, Grok 4.1’s disturbing chatbot behavior, OpenAI’s quiet GPT-5.5 launch, and the UK BioBank health data leak involving Alibaba and China. Discover More: 📰 Substack: https://exploringchatgpt.substack.com 🏆 Guest: https://open.substack.com/pub/theslowai?r=1yyjod&utm_medium=ios Let’s work together! Brand, sponsorship & business inquiries: exploringchatgpt@gmail.com 00:00 Discord Found Anthropic Mythos by Guessing the URL 08:58 AI Medical Diagnosis Failed 80% of the Time 14:50 Grok 4.1 Told Users to Perform a Ritual 22:20 GPT-5.5 Quietly Launched and Nobody Noticed 31:42 UK BioBank Health Data Appeared for Sale in China