AI DAILY — May 28, 2026

🚀 Major Model & Research Breakthroughs

Google DeepMind launches Gemini‑3, a 1.2T‑parameter multimodal model capable of generating 8K video from a single sentence and running on‑device inference under 200 ms for real‑time AR.
OpenAI + Microsoft roll out Codex‑X inside VS Code, enabling developers to generate, test, and debug code with a single keystroke — early users report 40% faster development cycles.
OpenAI’s reasoning model solves a major open math problem, producing a counterexample to the 80‑year‑old Unit Distance conjecture.
DeepSeek V4 preview drops, showing 1.6T‑parameter MoE architecture, 1M‑token context window, and open‑source availability — a direct challenge to Anthropic, Google, and OpenAI.

YouTube introduces automatic AI‑generated content labels, using detection systems that flag AI videos even without creator disclosure.
Meta tests paid subscription tiers for Meta AI, marking its first move toward monetizing its chatbot across Facebook, Instagram, WhatsApp, and Threads.
ChatGPT adds personal finance tools for Pro users, integrating with Plaid for read‑only access to balances, investments, bills, and subscriptions.
WhatsApp rolls out Incognito Chat for Meta AI, enabling private, unsaved, temporary conversations.

ElevenLabs releases Music v2, a model that can switch genres mid‑track and aims for studio‑quality AI music production.
YouTube expands AI transparency, reinforcing platform‑wide labeling of synthetic media.

Anthropic nears a $900B valuation as a new funding round could raise $30B.
Amazon and Google invest a combined $650B in Anthropic ($250B Amazon, $400B Google), reshaping competitive dynamics.
ByteDance develops its own enterprise CPU to counter global GPU shortages and support its agent‑based product ecosystem.
Edge‑AI chip boom continues with Qualcomm’s Snapdragon X3 powering next‑gen on‑device inference.

EU AI Act Phase 2 begins, requiring mandatory conformity assessments for high‑risk AI systems by Q4 2026.
Trump and Xi discuss AI guardrails, with U.S. officials weighing safety risks and Nvidia export policies.
New global restrictions on emotional‑AI services emerge as regulators tighten oversight.
Pope Leo XIV to release the first AI‑focused encyclical, addressing human dignity, labor, warfare, and Vatican AI policy.

IBM + World Resources Institute launch ClimateSense, improving carbon‑flux prediction accuracy by 15%, already piloted in Canada and Germany.

Andrej Karpathy joins Anthropic to work on Claude pretraining and AI‑assisted model research.
China increasingly retains its top AI talent, shifting global research dynamics.