March 23, 2026 – AI Daily Recap

From Yann LeCun’s record-shattering seed round to Cursor’s bid to dethrone frontier coding models, the U.S. Treasury stepping into AI governance, and OpenAI rolling ads into ChatGPT — Sunday was anything but quiet in the world of AI.

AMI Labs Raises $1 Billion in Europe’s Largest-Ever Seed Round

Advanced Machine Intelligence (AMI) Labs, founded by Turing Award winner and former Meta chief AI scientist Yann LeCun, has raised $1.03 billion in seed funding — the largest seed round in European history — at a $3.5 billion valuation. The Paris-based company hasn’t shipped a single product yet, but its backers include Nvidia, Bezos Expeditions, and Temasek. AMI Labs is focused on building “world models” for industrial and robotics applications, an approach that diverges from the large language model paradigm dominating most of the industry. The sheer size of the round signals investor conviction that the next frontier may lie beyond text generation.

Cursor Launches Composer 2, Its Own Frontier Coding Model

Cursor has released Composer 2, its own fine-tuned coding model built on Moonshot AI’s Kimi K2.5. The model features a 200,000-token context window, is tuned for multi-file edits, refactoring, and long-running agentic coding tasks, and costs roughly 86% less than its predecessor. Benchmarks show Composer 2 beating Claude Opus 4.6 on coding tasks, though GPT-5.4 still leads in some tests. With over 1 million daily active users and 50,000 business customers including Stripe, Cursor is betting it can compete on model quality — not just tooling.

U.S. Treasury Launches AI Innovation Series for Financial Sector

The U.S. Treasury Department’s AI Transformation Office (AITO) and the Office of the Financial Stability Oversight Council (FSOC) have launched the AI Innovation Series, a public-private initiative designed to support the U.S. financial system as AI adoption accelerates across banking, insurance, and capital markets. The program will convene financial institutions, regulators, and AI developers to address risks like model bias, systemic dependence on AI-driven trading, and the cybersecurity implications of automated financial infrastructure. It’s one of the clearest signals yet that the federal government is treating AI in finance as a stability concern, not just an efficiency play.

OpenAI Expands ChatGPT Advertising to All U.S. Free and Go Users

OpenAI will begin displaying advertisements to all U.S. users on the free and Go tiers of ChatGPT within weeks, partnering with ad-tech company Criteo as its first advertising technology partner. Early data from Criteo shows that users referred from LLM platforms convert at approximately 1.5x the rate of other referral channels. Advertisers are being asked to commit between $50,000 and $100,000 in initial spending. The move transforms ChatGPT from a pure utility into an ad-supported media platform — a shift that raises questions about how advertising incentives might influence the product’s behavior over time.

MiroThinker 72B Matches GPT-5 Reasoning — for Free

Miro Lab has released MiroThinker 72B, an open-source reasoning model that achieves 81.9% on GAIA benchmarks — comparable to paid GPT-5 performance on complex multi-step reasoning tasks. Built with an “interactive scaling verification” approach, the model delivers frontier-level logical reasoning without requiring a subscription or API fees. The release adds to a growing pattern: open-source models are closing the gap with proprietary systems faster than most anticipated, putting pressure on the pricing power of companies like OpenAI and Anthropic.

Kimi K2.5 Goes Global on Cloudflare’s Edge

Moonshot AI’s Kimi K2.5 is now available on the Cloudflare Workers AI platform, meaning developers can run a frontier-level model with a 256,000-token context window directly on global edge infrastructure — no server management required. The model is optimized for multi-step agentic tasks and tool calling, with strong visual understanding capabilities. This deployment model — combining a capable Chinese open-source model with Western cloud infrastructure — is becoming a recurring pattern that could reshape how AI compute is distributed globally.

The Big Picture

Sunday’s news reveals an AI industry rapidly fragmenting along new fault lines. LeCun’s AMI Labs is betting a billion dollars that the path to general intelligence runs through world models, not language. Cursor is proving that IDE-native companies can build competitive foundation models, not just consume them. The Treasury Department is quietly building the regulatory scaffolding for AI in finance, while OpenAI is embracing the advertising model it once seemed philosophically opposed to. And the continued rise of open-source challengers like MiroThinker 72B — delivering GPT-5-level reasoning for free — suggests the economic moats around proprietary AI may be thinner than their market caps imply.

Sources

• AI Releases March 23: Startups You Should Watch Now — Labla.org
• Cursor’s Composer 2 Beats Claude Opus 4.6 — VentureBeat
• Cursor Unveils Composer 2 for Cheaper AI Coding — WinBuzzer
• Treasury Launches the AI Innovation Series — U.S. Treasury
• OpenAI to Expand Ads to All Free and Go Users — The Tech Portal
• Criteo Joins OpenAI Advertising Pilot — Criteo
• AI News March 2026: In-Depth and Concise — The AI Track
• Cursor Composer 2: Frontier Agentic Coding Model Debuts — AI CERTs