The dust is still settling from OpenAI's Astra for Law launch, and today's stories clarify what that move actually did: it turned a market of vertical legal AI vendors into a platform competition, and most of those vendors were building on OpenAI's own infrastructure. The battle for 'centrality' — which system becomes the default workflow layer for lawyers — is now explicitly a fight between OpenAI's platform ambitions and whoever has a durable data moat or workflow lock-in. Harvey has its valuation. Legora has its citator. Google has its enterprise relationships and NotebookLM. None of that is irrelevant, but the question of who owns the lawyer's primary interface just got a lot harder.

Meanwhile, a Deloitte survey of 25,000 workers finds that AI productivity gains have been slim for most of the workforce — and the legal industry is asking whether it's the exception. It probably isn't, but the question matters: 31% of law firms say AI improved profitability, which means 69% either didn't measure or didn't find the gains. The legal AI market is generating a lot of press releases and not enough P&L line items. The vendors who can close that gap are going to win; everyone else is selling workflow theater.

The US and China are reportedly discussing a mutual alerting mechanism for AI-related national security threats — a confidence-building measure that would be the first structured bilateral AI safety channel between the two countries. This is worth watching without getting too excited about. A hotline for AI incidents is not a governance framework, a capability agreement, or anything close to enforceable. What it does signal is that both governments have acknowledged the category of 'AI surprises with national security consequences' as real enough to warrant communication infrastructure. That's a floor, not a ceiling, and the ceiling remains very far away.

The broader slowdown debate — still the loudest conversation in AI circles — continues to produce more rhetoric than mechanism. The Trump administration's posture remains that voluntary restraint is the labs' own problem. Jensen Huang has been clear about his position. And China has been clear that it views any Western coordination as an opportunity rather than a norm. What's left of the slowdown argument is a sincere and growing coalition with no enforcement architecture and no geopolitical alignment. That's not nothing, but it's not policy.

Google has launched the Googlebook at $899 — a laptop whose core value proposition is that it runs Gemini better than your current machine. That is either a smart vertical integration play or a $899 bet that consumers will buy new hardware for an AI assistant they're currently accessing for free from any browser. Apple's M6 Mac Mini is getting reviewed in the same week as 'perfect for the AI curious,' which is a gentler version of the same pitch. The hardware-as-AI-distribution-channel thesis is getting its first real commercial test, and the answer will be visible in sales data well before anyone writes the retrospective.