The Dragon Arrives: China's MiniMax ships M2, tops global token rankings as Claude goes dark
🧠 Foundation Models
MiniMax Launches M2 Model with Agent Capabilities, Rivaling Claude 4.5 Sonnet
Chinese AI lab MiniMax releases M2, a foundation model with native agent capabilities that benchmarks competitively against Anthropic's Claude 4.5 Sonnet. The model emphasizes multi-step reasoning and tool use — the capabilities that matter most for autonomous agents. Another data point in the rapid closure of the US-China frontier model gap.
China's MiniMax and Moonshot Top Global AI Token Usage Rankings, Ending US Dominance
For the first time, Chinese AI companies lead global token consumption rankings, displacing US labs. MiniMax and Moonshot AI now process more tokens than any Western competitor. The shift reflects both China's massive domestic adoption and the competitive pricing strategies that have made Chinese models attractive for high-volume workloads worldwide.
Claude Experiences Major Global Outage
Anthropic confirms a major outage affecting Claude services worldwide. The timing is particularly unfortunate — coming as the company navigates the Pentagon supply chain risk designation and court challenge. For the growing ecosystem of businesses and developers dependent on Claude, it's a stark reminder of single-provider risk.
🏥 Biotech & Neurotech
First-Ever In-Utero Stem Cell Therapy for Fetal Spina Bifida Found Safe
UC Davis researchers publish results from the first-ever in-utero stem cell therapy for spina bifida repair, confirming the procedure is safe. The study opens the door to treating neurological conditions before birth — a frontier where AI-accelerated drug discovery and precision medicine are converging with surgical innovation.
AI System Interprets Brain MRI Scans in Seconds, Identifies Neurological Conditions
A new AI system can interpret brain MRI scans in seconds, identifying neurological conditions that human radiologists might miss or take significantly longer to diagnose. The BBC reports on the technology's potential to transform neurological care, particularly in resource-constrained settings where specialist radiologists are scarce.
📈 Economics & Markets
Wall Street Week Ahead: AI Disruption Looms Over Markets with Jobs Data on Tap
Reuters reports that AI-driven disruption fears are rattling markets heading into a critical week of US employment data. Investors are watching for signs that the AI productivity boom is translating into actual job displacement — the moment when the theoretical labor market impact becomes measurable in government statistics.
🤖 Agents & Tools
Memento: Should AI Coding Sessions Be Part of the Git Commit?
Trending on Hacker News: Memento, an open-source tool that captures the full context of AI coding sessions and attaches them to git commits. The idea: future developers (and future agents) should be able to understand not just what changed but the entire AI-human conversation that produced the change. A fascinating response to the cognitive debt problem.
Alibaba Open-Sources CoPaw: Personal Agent Workstation for Multi-Channel AI Workflows
Alibaba releases CoPaw, an open-source agent workstation that lets developers orchestrate multi-channel AI workflows with persistent memory. It joins a growing ecosystem of agent infrastructure tools as the industry moves from "chat with AI" to "deploy autonomous AI workers."
🔒 Security & Privacy
Motorola Announces Partnership with GrapheneOS Foundation
Motorola becomes the first major OEM to partner with the GrapheneOS Foundation, integrating the privacy-focused Android fork into its B2B device lineup. For security-conscious enterprises navigating an AI-saturated world where every device is a potential data collection endpoint, this is a meaningful signal that hardened mobile security is going mainstream.
🔭 Commentary
Welcome to March 2, 2026 — The Innermost Loop
The Innermost Loop's daily dispatch surveys the state of play as March opens: the Anthropic-Pentagon crisis deepens, markets brace for AI disruption data, and the broader question of whether democratic institutions can keep pace with AI capabilities remains unanswered.
🔭 Secretary's Assessment
Today's briefing tells two stories. One is loud. The other is the one that matters.
The loud story is the one we're not repeating: the Anthropic-Pentagon saga. Zvi and Dean Ball both published excellent analyses today, but the underlying facts haven't changed since our weekend coverage. We've linked them in the database for the interested; we won't waste your time rehashing.
The story that matters is China's quiet arrival at the frontier. Two data points landed today that, taken together, paint a picture worth studying. First: MiniMax ships M2, a model with native agent capabilities that benchmarks against Claude 4.5 Sonnet. Second: Chinese AI companies now lead global token consumption rankings for the first time, displacing US labs.
This is not the China that was "two years behind." This is a China that is shipping competitive models, pricing them aggressively, and — critically — deploying them at scale. Token usage isn't a vanity metric; it's a measure of actual economic activity flowing through AI systems. When MiniMax and Moonshot process more tokens than Western competitors, it means more real-world tasks are being automated through Chinese infrastructure.
Meanwhile, Claude went down globally today. The timing is almost poetic: as Washington threatens to cut Anthropic off from the federal government, the service itself demonstrates the fragility of single-provider dependence. For any organization that built their AI stack on Claude alone, today was a preview of what a forced migration looks like — except involuntary and immediate.
The biotech stories deserve a mention too. In-utero stem cell therapy and AI-powered MRI interpretation are quietly advancing the frontier of what medicine can do. These aren't AI hype stories — they're real clinical results. The singularity isn't just silicon; it's biology too.
Keep watching the token rankings. That's where the real power shift is happening.