Friday Edition

AI News Roundup — May 21, 2026

Google I/O 2026 dominated the day with Gemini 3.5 Flash, Gemini Omni, and 100 major announcements — including AI Mode crossing 1 billion monthly users and the biggest Search redesign in 25 years. Meanwhile, OpenAI cracked an 80-year-old mathematical problem and the Pentagon began testing Claude replacements.

Google I/O 2026: Gemini 3.5 Flash, Gemini Omni, and AI Mode Hits 1 Billion Users

Google's annual developer conference delivered 100 announcements. Gemini 3.5 Flash is now generally available — it rivals large flagship models at Flash speeds, topping Terminal-Bench 2.1 (76.2%), GDPval-AA (1656 Elo), and MCP Atlas (83.6%). Gemini Omni is a new model that creates any output from any input, starting with video — it understands physics (gravity, kinetic energy, fluid dynamics) and is available now in Google Flow, YouTube Shorts Remix, and the Gemini app. AI Mode has crossed 1 billion monthly users and is being upgraded with Gemini 3.5 Flash as its default model. The Search box gets its biggest redesign in 25 years, accepting text, images, files, videos, and Chrome tabs simultaneously. Gemini 3.5 Pro is already in internal testing and expected to roll out next month.

Google Blog · Sundar Pichai Keynote


OpenAI Cracks an 80-Year-Old Erdős Mathematics Problem

OpenAI's reasoning models have solved a famous unsolved problem originally posed by Hungarian mathematician Paul Erdős — a new high-water mark showing AI systems surpassing human experts in formal mathematical reasoning at the frontier.

The Guardian


Pentagon Tests Rival AI Models to Replace Anthropic's Claude

The Department of Defense is evaluating AI models using 25 power users to find a replacement for Claude after the federal blacklisting of Anthropic. OpenAI, xAI, and Google are all competing. Scientific American reports the transition could take months due to the complexity of classified deployment pipelines.

Bloomberg · Scientific American


Google Antigravity 2.0: Agentic Development Platform Ships as Standalone App

Google Antigravity 2.0 is a new standalone desktop application for orchestrating multiple AI agents in parallel — one agent can code a website while another generates brand assets simultaneously. The Antigravity CLI offers terminal-native agent creation. Antigravity is now embedded in Search, enabling custom generative UIs and persistent mini-apps built on the fly.

Google Antigravity Blog


Trump Postpones Signing AI Executive Order; Anthropic Raises Weekly Usage Limits

President Trump delayed signing a major AI executive order that was expected to shape federal AI procurement rules. Separately, Anthropic raised weekly usage limits by 50% until July 13 for existing subscribers, as a goodwill measure amid the ongoing federal contracting freeze.

PBS NewsHour


Ask YouTube Launches; SynthID Reaches 50 Million Verifications Globally

Google launched Ask YouTube, a conversational search experience that compiles answers from across YouTube's entire catalog including long-form videos and Shorts. Rolling out this month to a subset of US English desktop users. SynthID, Google's AI watermarking technology, has now been used 50 million times globally — verification is expanding to Search and Chrome.

YouTube Blog


Search Agents Arrive: Google's Background Information Agents Launching This Summer

Google announced "information agents" — persistent AI agents that run 24/7 in the background monitoring any topic, task, or project, then deliver synthesized updates with the ability to take action. Users can run multiple agents simultaneously. Rolling out first to Google AI Pro and Ultra subscribers this summer.

Google Search Blog