AI News Roundup: Anthropic Sues Pentagon, Meta Buys AI Bot Network, NVIDIA GTC Preview

Anthropic Sues the Pentagon Over Supply-Chain Risk Label

In the biggest AI-policy story of the week, Anthropic filed suit against the U.S. Department of Defense on Sunday after the agency designated it a “supply-chain risk.” The dispute erupted when contract negotiations broke down over two red lines Anthropic refused to cross: allowing its AI to be used for mass surveillance of American citizens, or for autonomous weapons systems.

The lawsuits, filed in both a California district court and the D.C. federal appeals court, allege the Trump administration violated Anthropic’s First Amendment rights and overstepped the scope of supply-chain risk law. The designation could jeopardize hundreds of millions of dollars in government revenue for the AI company.

In a notable show of cross-company solidarity, more than 30 employees from OpenAI and Google DeepMind filed an amicus brief in their personal capacities supporting Anthropic. Microsoft also filed its own amicus brief in support. Meanwhile, Google has been quietly expanding its own Pentagon work while its rivals publicly spar over the conditions for Defense Department contracts.

Meta Acquires Moltbook, the Social Network for AI Agents

Meta has acquired Moltbook, the Reddit-like platform where AI agents can post, comment, and vote on content while their human creators watch from the sidelines. The deal, first reported by Axios, brings Moltbook CEO Matt Schlicht and COO Ben Parr into Meta’s Superintelligence Labs unit.

Moltbook became the talk of Silicon Valley after racking up millions of registered bots within days of its launch. However, security researchers revealed that the platform had significant vulnerabilities, making it easy for humans to impersonate AI agents. The acquisition comes just weeks after OpenAI hired the founder of OpenClaw, the technology that powered Moltbook’s agent interactions.

NVIDIA GTC 2026: Jensen Huang Promises a Chip That Will “Surprise the World”

All eyes are on San Jose as NVIDIA GTC 2026 approaches next week (March 16–19). CEO Jensen Huang will deliver his keynote on Monday, March 16 at 11 a.m. PT, with more than 30,000 attendees from 190 countries expected. Huang has teased an unidentified new chip that he says will “surprise the world,” though NVIDIA hasn’t confirmed whether it’s a data-center accelerator, networking silicon, or something else entirely.

The conference will cover the full AI stack—from accelerated computing and AI factories to agentic systems, robotics, digital twins, and quantum computing. The keynote will be livestreamed free at nvidia.com.

Google Rolls Out Gemini 3 Canvas to All U.S. Users

Google expanded Canvas in AI Mode to all U.S. users, bringing coding, writing, and app-generation capabilities directly into Google Search. Powered by the new Gemini 3 model, Canvas lets users draft documents, generate code, and build custom tools without leaving the search interface.

Beyond Search, Google announced a broad expansion of Gemini-powered AI across Docs, Sheets, Slides, and Drive, while introducing new user controls for AI features in Google Photos. Gemini 3 is also now the default model for AI Overviews globally, meaning AI-generated responses will appear on the search results page for more queries worldwide.

OpenAI Ships GPT-5.4 With 1 Million Token Context Window

OpenAI released GPT-5.4 on March 5, combining improved coding and reasoning capabilities with a massive 1,000,000-token context window. The release keeps OpenAI competitive in the context-length race—Anthropic’s Claude Sonnet 4.6 also offers a 1-million token context window since its February launch.

China’s AI Race Heats Up

Five new AI models dropped from China’s top contenders in recent weeks, with entries from Tencent, Alibaba, Baidu, and ByteDance. The standout is MiniMax’s M2.5, which benchmarks claim rival Anthropic’s Claude Opus 4.6 at a fraction of the cost—a significant development in the ongoing price war between Chinese and Western AI providers.