Grant Harvey

Microsoft open-sourced ECHO; Odyssey launched Starchild-1 and Agora-1; OpenAI added bank-account connections to ChatGPT; Nvidia GPU rentals spiked; AI backlash widened across polls, jobs, data centers, markets, and surveillance fights.

Anthropic refused China access to Mythos while the Pentagon used the model family for cyber defense; Cerebras and Cowboy Space pushed AI compute into new territory; Recursive Superintelligence raised $650M; Google confirmed criminal AI-driven zero-day exploitation; plus the week's best new tools.

Recursive Superintelligence raised major funding to build self-improving AI; Nous Research claimed 2-3x faster LLM pretraining; NVIDIA shipped elastic reasoning and any-step video models; Kimi launched browser-control agents; Anthropic's CFO explained the compute economy.

Anthropic refused China access to its newest model; Isomorphic raised $2.1B; Google pushed Gemini deeper into Android; xAI shipped Grok Voice Think Fast; supply-chain attackers hit Mistral and TanStack.

Google’s Magic Pointer, Perceptron Mk1, and Thinking Machines’ interaction models all point toward the same shift: AI interfaces are moving from “type a perfect prompt” to “show the system what you mean.”

Cerebras upsized its IPO to $4.8B at $33B as Cowboy Space raised $275M to put data centers in orbit; Google confirmed the first criminal AI-discovered zero-day; OpenAI launched DeployCo with $4B+; Anthropic shipped Claude Platform on AWS; METR engineers self-reported 1.4-2x AI value; Mira Murati's Thinking Machines unveiled real-time interaction models.

Anthropic is weighing a $50B primary round at a $900B valuation that would top OpenAI; the Trump administration drafted an AI security order that omits mandatory model tests; Apple and Intel reached a Trump-pushed preliminary chip-making agreement; French prosecutors escalated their Musk and X probe to a criminal investigation; the US suspects a Thai national-AI partner helped smuggle Nvidia chips to Alibaba; Cerebras is set to raise its IPO range on 20× oversubscription.

Microsoft surveyed 20,000 AI users and analyzed trillions of Microsoft 365 signals for its 2026 Work Trend Index. The headline isn't that AI works; it's that most organizations aren't built to capture what their workers can already do.