The biggest AI story in today's feed is one you've already seen: OpenAI's Astra for Law launched yesterday, and today the legal tech press is still processing it. We're not re-running the launch — that's in your previous briefing — but the downstream signal is worth noting: Meta is in a similar position in the consumer space, where its Muse platform is drawing scrutiny that the launch press didn't. Wired's hands-on review finds Muse better at passive surveillance of your behavior than active assistance with your problems, which is either a feature or a bug depending on whether you're a user or an advertiser.

Meanwhile, Trump is floating an AI rebranding exercise and standing up something called an 'AI Force,' which tells you everything you need to know about where federal AI policy is right now: long on nomenclature, short on enforcement architecture.

OpenAI's GPT-6 Astra has apparently come for a human's mathematical world record — specifically in twin primes research, per Science News. The human set a record; the AI came for it shortly after. Whether this is a genuine research contribution or a very fast pattern match on known methods is exactly the question the mathematics community is now arguing about, which is itself an interesting signal: AI math claims are no longer abstract enough to ignore, and the dispute is happening among people who can actually check the work.

Stanford HAI adds a useful adjacent concern: world models — the internal simulations that frontier AI systems use to reason about physical and social environments — are becoming capable enough that governing them is now a distinct policy challenge, not a subset of the general model governance conversation.

Below the headline launches, the more durable story today is that AI is getting bolted into specific professional workflows with real institutional backing. Husch Blackwell has shipped CXT, an AI-native case management platform for complex torts defense — not a chatbot wrapper, but a concierge operational model. The MIT-IBM Computing Research Lab is moving quantum and AI deployments from theory to production systems. And Meta's Muse Spark 1.1 is quietly shipping as an API product, which matters more for developers than the consumer Muse reviews suggest. The pattern: vertical AI is past the proof-of-concept phase and into the 'does it actually run the workflow' phase, which is where most enterprise technology either earns its keep or disappears.