AI Daily Brief

Thursday, May 21, 2026

7 stories · Karpathy joins Anthropic, Google I/O wave, and the race for AI talent

Today's Stories

Andrej Karpathy joins Anthropic's pre-training team

OpenAI co-founder and former Tesla AI director Andrej Karpathy has joined Anthropic, signing on to lead a new research unit within the pre-training division. His team's mandate is to use Claude to autonomously accelerate pre-training research — a form of recursive self-improvement. The hire is widely seen as a coup: Karpathy is one of the most respected names in deep learning, and his choice of Anthropic over a return to OpenAI signals shifting gravity in the frontier AI race.

Read on TechCrunch →

Anthropic's valuation surges past $1 trillion as $30B round nears close

Anthropic's annualised revenue has tripled to $30 billion since December, and the company's secondary-market valuation has already eclipsed $1 trillion. A formal $30 billion funding round at roughly $900 billion valuation is in final stages — with Google and Amazon already having committed $40B and $5B respectively, packaged alongside $100B of committed AWS spend. The company is now targeting a public offering in the second half of 2026.

Read on Fortune →

Google I/O 2026: Gemini Spark, Omni, Antigravity 2.0, and the agentic era

Google used I/O 2026 to go all-in on agentic AI. Gemini 3.5 Flash — four times faster than peer frontier models — leads a new series built for coding and agentic workflows. Gemini Omni is a new multimodal model that generates outputs in any modality from any input, including video. Gemini Spark is a persistent AI agent that lives on a dedicated virtual machine in Google Cloud, can manage your calendar, email, Drive files, and proactively act on recurring tasks. Developers get Antigravity 2.0 — an upgraded agentic dev platform — plus a proposed open web standard called WebMCP for browser-based agents.

Read on blog.google →

Google cuts AI Ultra to $200/mo, adds $100 entry tier in pricing reset

Alongside the I/O announcements, Google trimmed the price of its top-tier AI Ultra subscription from $250 to $200 per month — a 20% cut — and introduced a new $100-per-month entry-level AI Ultra tier, broadening access to its most powerful tools. The moves follow aggressive pricing pressure from OpenAI and Anthropic and appear designed to pull prosumers and developers off rival platforms rather than chase the mass-market segment.

Read on CNBC →

OpenAI Codex lands in ChatGPT mobile — and takes direct aim at Claude Code

OpenAI launched Codex integration inside the ChatGPT mobile app (iOS and Android) on May 14, letting developers monitor and manage multi-environment coding workflows from their phones. The May update also added quick TUI reasoning controls, first-class Amazon Bedrock support, a remote plugin marketplace, and stable MCP and Bash hooks. Meanwhile Claude Code added a new "claude agents" dashboard for session management and the /goal command for persistent autonomous runs — keeping the two platforms in a tight feature race.

Read on The New Stack →

xAI launches Grok Connectors, integrating SharePoint, Outlook, Google Workspace, and more

xAI released Grok Connectors in Grok Web, adding deep integrations with everyday productivity tools: SharePoint, Outlook, OneDrive, Google Workspace, Notion, GitHub, and Linear. A new Grok Build beta also opened to SuperGrok Heavy subscribers, offering an interactive TUI, headless scripting mode, and Agent Client Protocol support. Separately, xAI added Quality Mode to the Grok Imagine API, promising higher realism and stronger text rendering for enterprise image generation workflows.

Read on x.ai →

Microsoft, Google, and xAI agree to share pre-launch AI models with the US government

Three of the biggest names in frontier AI — Microsoft, Google, and xAI — have agreed to share unreleased versions of their models with the Center for AI Standards and Innovation (CAISI) for pre-launch evaluation. The arrangement targets cybersecurity risks, giving federal researchers early access before public release. The move follows ongoing pressure from regulators and is viewed as a way for the companies to get ahead of mandatory disclosure rules while building goodwill with the incoming AI governance framework.

Read on CNN Business →