OpenAI is officially breaking away from its total reliance on standard Nvidia hardware by unveiling its very first custom-designed AI inference processor, developed alongside Broadcom. At the exact same time, Google has struck back in the agent race by natively baking advanced "computer use" capabilities right into its mainstream Gemini 3.5 Flash model.

We’ll talk about:

• OpenAI pulling back the curtain on "Jalapeño," a custom Broadcom-manufactured inference chip designed with the help of OpenAI's own AI models.
• Google Gemini 3.5 Flash shipping with native, built-in computer use and advanced enterprise guardrails to block indirect prompt injections.
• Anthropic publicly accusing Alibaba of a massive, brazen data-scraping attack using 25,000 fake accounts to train its own models.
• The gr ...