28th November - AI News Daily - OpenAI Confirms Data Breach, Severs Vendor Ties Amid Security Overhaul

Send us a text 🌍 INAI • The Open AI Hub The Intelligence Atlas → the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news —free for all, updated every day. https://github.com/inai-sandy/inAI-wiki Top Highlights: Google launched Gemini 3, advancing agentic automation and multimodal reasoning. OpenAI confirmed a Mixpanel data breachand expanded data residency to 10 regions. Publishers filed a $10B copyright suit against OpenAI and Microsoft. DeepSeek-Math-V2 achieved IMO gold-level performance with open weights. An MIT study found 11–12% of U.S. jobs are automatable by agentic AI, while CEOs warned of power and data center shortages. New Tools: LangChain Deep Agents launched secure sandboxes for code execution. n8n + MCP integrations enable ChatGPT and Claude to manage workflows. fal FLUX.2 shipped to production, vLLM Ray APIs streamlined MoE inference, Google Antigravity IDE debuted with Gemini 3, and OpenAI Sora rolled out on Android globally. LLM Updates: Google Gemini 3 strengthened language, vision, and reasoning. DeepSeek-Math-V2 posted IMO gold performance. INTELLECT-3 scaled RL to 100B+ MoE. Microsoft Fara-7B targeted computer-use tasks. Anthropic Claude Opus 4.5 added design iteration and PC control demos. Benchmarks showed Claude Opus 4.x edging Gemini Pro 3 on scientific tasks. Research: JECS GWAS dataset released childhood genetics data. Sparse attention gained recognition, mean-pooling emerged as strong compression. Beluga CXL memory cut LLM latency ~90%. P1 physics model achieved gold at International Physics Olympiad. MIT estimated 11–12% of jobs automatable. Industry: OpenAI ended Mixpanel partnership after breachand added 10-region data residency. News outlets sued OpenAI and Microsoft for $10B. UK AISI allocated £15M for alignment research. Nadella and Altman warned of energy constraints. Researchers flagged security gaps in popular stacks. Tutorials: Anthropic shared agent guidance, Hugging Face explained modern inference, multi-vector retrieval deep dives covered RAG optimization, builder summits addressed quantization, and a Deep Representation Learning course released materials. Showcases: Anthropic's Eiffel Tower Llama demo recreated, LLMs roleplayed Mafia, Claude Opus 4.5 showed PC control, Flux 2 launched art contest, and Colombian students won NASA Space Apps Challenge. Discussions: Ilya Sutskever argued breakthroughs need original research. Governance debates intensified. Enterprises favored open AI. Studies highlighted idea diversity benefitsand agent reliability challenges. Support the show 🌍 INAI • The Open AI Hub The Intelligence Atlas → the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news —free for all, updated every day. https://github.com/inai-sandy/inAI-wiki