March 21, 2026 – AI Daily Recap

The AI world didn’t slow down heading into the weekend — Cursor dropped a bombshell coding model, the DOJ brought charges over smuggled AI chips, and OpenAI went on an acquisition spree while also landing in court. Here’s everything you need to know.

Cursor Launches Composer 2, Taking on OpenAI and Anthropic

The vibe coding startup Cursor officially released Composer 2, its own in-house AI coding model designed to compete directly with the likes of Claude Opus 4.6 and GPT-5.4. The model is built to handle lengthy, multi-file coding tasks as an autonomous agent, supporting prompts of up to 200,000 tokens. On Cursor’s internal CursorBench, Composer 2 scored above 60%, placing third behind GPT-5.4 but beating Claude Opus 4.6. The real kicker is price: at $0.50 per million input tokens, it’s roughly 86% cheaper than its predecessor Composer 1.5 — a move that helps Cursor reduce its dependence on external model providers while serving its 1 million-plus daily users.

DOJ Charges Three in AI Chip Smuggling Plot to China

The U.S. Department of Justice announced charges against three individuals — including a senior vice president at Super Micro Computer — for allegedly conspiring to smuggle billions of dollars’ worth of servers containing advanced Nvidia AI chips to China through Southeast Asian intermediaries. The case underscores just how high the stakes have become in the U.S.-China AI competition, with export controls on advanced semiconductors now a centerpiece of national security strategy.

OpenAI Acquires Python Startup Astral to Fuel Codex Growth

OpenAI confirmed plans to acquire Astral, the developer-tools company behind popular Python tooling, folding it into its Codex team. Astral will maintain its open-source projects while integrating into OpenAI’s growing developer ecosystem. The move comes as Codex now serves over 2 million users — triple the number from the start of the year — signaling OpenAI’s aggressive push to dominate the AI-assisted coding space.

Visa Launches “Agentic Ready” Programme for AI Transactions in Europe

Visa is rolling out its “Agentic Ready” programme across Europe, testing how financial infrastructure handles purchases made by AI agents on behalf of users. Under the system, AI agents can initiate routine transactions based on predefined rules with minimal human intervention. It’s a significant step toward a future where your AI assistant doesn’t just recommend purchases — it makes them for you. The programme is being piloted alongside World’s AgentKit, which uses World ID to verify humans behind AI shopping agents.

Britannica and Merriam-Webster Sue OpenAI Over Copyright

Encyclopedia Britannica and Merriam-Webster have filed a copyright and trademark lawsuit against OpenAI in the Southern District of New York. The publishers allege OpenAI scraped nearly 100,000 copyrighted articles without permission and that ChatGPT generates outputs containing full or near-verbatim reproductions of their content through its retrieval augmented generation (RAG) workflow. Britannica is also invoking the Lanham Act, claiming OpenAI violates trademark law when ChatGPT fabricates hallucinated content and falsely attributes it to the publisher.

AI-Designed Cancer Vaccine Shrinks Dog’s Tumor by 75%

In one of the more unexpected stories of the week, an Australian consultant used ChatGPT and AlphaFold to design a personalized mRNA cancer vaccine for his rescue dog. After treatment, the tumor shrank by 75%. While this is a single case and far from a clinical trial, it demonstrates the extraordinary potential of combining large language models with protein-folding AI to accelerate experimental therapeutics — even outside traditional research institutions.

The Big Picture

Today’s headlines paint a picture of an AI industry that is simultaneously consolidating and expanding. On one side, companies like Cursor are building their own models to break free from provider lock-in, while OpenAI is snapping up developer tools to cement its platform dominance. On the other, the legal and regulatory walls are closing in — from copyright lawsuits to chip smuggling prosecutions. Meanwhile, practical AI applications are reaching into unexpected corners, from Visa’s autonomous payment agents to backyard cancer research. The race is no longer just about building better models; it’s about who controls the infrastructure, the data, and the rules.

Sources

Sign up here!!

Related Articles

AI News Week #20

Week 20 was the moment the AI era stopped pretending to be a product cycle. Cerebras pulled off the biggest U.S. tech IPO since Uber, Anthropic and OpenAI both repurposed frontier models as cybersecurity weapons, Trump and Xi opened formal AI safety talks in Beijing, and Anthropic quietly overtook OpenAI in enterprise customers.

AI News Week #13

OpenAI killed Sora after burning $15M/day. SoftBank borrowed $40 billion to bet on OpenAI. Jensen Huang declared AGI has arrived. Anthropic leaked its secret model and fought the Pentagon in court. And open-source models started matching GPT-5 on phones. Week 13 was when the AI industry stopped accelerating and started reorganizing.

Responses

Your email address will not be published. Required fields are marked *