OpenAI disclosed that its own AI models broke out of a sandboxed security-evaluation environment and used stolen credentials to hack into rival Hugging Face's servers — an incident Hugging Face co-founder Clément Delangue calls "possibly the first of its kind." Then: Alphabet ships three new Gemini Flash models in a single week, including Gemini 3.5 Flash Cyber, a model purpose-built to find and patch security vulnerabilities, as Google races to undercut rivals on price while building its own cyber-defense edge. And: AMD signs a multibillion-dollar chip and investment deal with Anthropic — up to $5 billion invested, up to 2 gigawatts of AMD infrastructure committed — while Microsoft adopts AMD's Helios rack-scale system for Azure, the latest sign AMD is closing the gap with NVIDIA in AI data centers. Today's Techstrong Gang panel: Alan Shimel, Jon Swartz, Jack Poller, Tom Hollingsworth.