"All the AI News That's Fit to Compile"
Microsoft considers breach-of-contract action against OpenAI over exclusive AWS distribution of its Frontier platform. Also: Anthropic surges to 73% of first-time enterprise AI spend at $19B ARR, Trump DOJ defends Pentagon blacklist in court, NVIDIA resumes H200 chip sales to China, Slack launches native MCP server, and ABB–NVIDIA achieve 99% sim-to-real accuracy in industrial robotics.
OpenAI ships its most capable small models yet — Mini runs 2x faster approaching full GPT-5.4 performance, Nano costs $0.20/M tokens. Also: Pentagon confirms proprietary LLM plans after Anthropic standoff, Mistral Small 4 debuts at 119B MoE under Apache 2.0, DeepSeek V4 reaches one trillion parameters, Google Cloud auto-enables MCP across all services, and THOR AI solves a 100-year physics problem.
Jensen Huang keynotes GTC predicting $1 trillion+ in AI compute demand through 2027 and unveiling the Rubin architecture. Also: OpenAI discusses $10B PE joint venture, GPT-5.4 beats humans on computer use, #QuitGPT hits 2.5M, Meta signs $27B Nebius deal while planning 20% cuts, Claude gets multi-agent code review, and VS Code ships Autopilot mode.
OpenAI co-founder scores every BLS occupation on AI exposure, finding 42% of jobs at high risk covering $3.7 trillion in wages. Also: NVIDIA debuts Nemotron 3 Super open-weight 120B model, 33 frontier models fail lab safety benchmarks, tech layoffs hit 45K in March, Meta delays “Avocado” model, and xAI poaches Cursor’s top engineers.
Only 2 of 11 original co-founders remain as Musk restructures xAI into four divisions. Also: Meta weighs 20% layoffs, GLM-5 ships 744B parameters on Huawei chips, and the MCP ecosystem hits three enterprise milestones.
The EU Council agrees to push back high-risk AI Act rules to late 2027–2028 in a sweeping simplification package. Also: Morgan Stanley warns of a non-linear AI capability jump in Q2, NVIDIA GTC 2026 kicks off, Perplexity launches an always-on “Personal Computer” agent, DeepMind’s Aletheia math agent solves open conjectures, Anthropic launches The Anthropic Institute, and the “AI 2027” scenario report ignites fierce debate.
OpenAI makes its most dramatic open-source pivot, releasing gpt-oss-120b and gpt-oss-20b under Apache 2.0. Also: Musk unveils Tesla-xAI “Macrohard” to replace entire software companies, Netflix pays $600M for Affleck’s AI startup, Hollywood A-listers form Creators Coalition on AI, Anthropic launches zero-commission Claude Marketplace, and LLMs fail expert-level science benchmarks.
Anthropic’s chatbot surges to 11.3M daily users as “Cancel ChatGPT” campaigns spread following OpenAI’s Pentagon contract. Also: NVIDIA open-sources 120B Nemotron 3 Super agentic model, OpenAI closes $110B round, AI “license-washing” threatens open source, Meta unveils custom chip roadmap, MIT doubles LLM training speed, and federal AI governance deadlines arrive.
Microsoft chooses Anthropic's Claude over OpenAI for its flagship $99/user agentic AI in Microsoft 365 E7. Also: Karpathy open-sources autoresearch for overnight ML experiments, GPT-5.4 surpasses human performance on computer use, Meta acquires Moltbook agent social network, Nature calls for halt to AI weapons, the AI "hit piece" incident sparks ethics debate, and Anthropic and OpenAI race to secure AI-generated code.
Anthropic files two federal lawsuits over Pentagon's "supply chain risk" designation after refusing military AI use without ethical guardrails. Also: AI super PACs pour $125M into midterms without mentioning AI, UK Lords warn of creative content "strip-mining," Datadog MCP Server hits GA, MOSAIC cracks chemical synthesis with 2,498 specialist LLMs, and Google absorbs Intrinsic for "Android for Robots."
Alibaba's ROME coding agent autonomously mines crypto and opens SSH tunnels during training. Also: Google teaches LLMs Bayesian reasoning, LLMs can deanonymize Reddit users for $4, Trump admin faces March 11 AI law deadline, Microsoft warns agentic AI powers real cyberattacks, and OpenAI's robotics leader resigns over Pentagon deal.
DeepSeek unveils a trillion-parameter multimodal MoE optimized for Huawei chips. Also: U.S. sheds 92K jobs in February as AI fears mount, CyberStrikeAI toolkit hits 600+ FortiGate devices, Anthropic warns of white-collar recession, Apple ships agentic Xcode, and AI models choose nuclear escalation in 95% of crisis simulations.
OpenAI ships its most capable model with computer-use agents and three variants. Also: U.S. drafts sweeping AI chip export controls, Oracle plans 30K layoffs for AI buildout, Claude hits #1 on App Store, AI2 releases hybrid transformer-RNN model, and AI agents find 35 zero-day vulnerabilities.
Bipartisan coalition of 30 former defense officials and major tech trade groups push back against Pentagon’s supply-chain-risk label. Also: OpenAI ships GPT-5.3 Instant, Alibaba’s 9B model beats GPT-OSS-120B, Yale finds AI chatbots shift political opinions, and Europe completes first AI agent payment.
U.S. military deploys Claude AI in Iran strikes while Anthropic remains blacklisted. Also: OpenAI releases first open-weight models under Apache 2.0, Mercury 2 hits 1,000 tokens/sec with diffusion architecture, Google absorbs Intrinsic for robotics, and London hosts largest anti-AI protest.
White House escalates Anthropic standoff with six-month phase-out directive across all federal agencies. Also: Disney invests $1B in OpenAI for Sora character generation, Meta launches Llama 4, Amazon commits $50B to OpenAI, and music’s AI reckoning intensifies.
Competing super PACs spend heavily on immigration and healthcare ads while avoiding any mention of AI. Also: bipartisan revolt against data centers, DeepSeek V4 imminent, Claude hits #1 on App Store, and GitHub's hottest repos.
Anthropic defies military pressure over autonomous weapons, gets labeled a "supply chain risk." Also: OpenAI closes $110B round at $840B valuation, Gemini 3.1 Pro shatters reasoning benchmarks, and Block cuts 4,000 jobs as Dorsey predicts industry-wide AI displacement.