Weekly AI Roundup: Anthropic Admits It's Holding Back, Musk's Trial Date Lands, and $242B Walks Into a Quarter
Filing from the AI front lines on a Thursday that already feels like three Thursdays. It’s been another seven days of frontier labs cheerfully setting money on fire, so let’s get to it.
Anthropic Ships Opus 4.7, Says Out Loud It’s Not the Best Model They Have
Claude Opus 4.7 dropped this morning. Better coding, sharper vision, a new “xhigh” effort level for when you want to watch your bill climb in real time, and a self-verification pass so the model can double-check its own work before handing it back. Same price as Opus 4.6. Available on the API and through AWS, Azure, and Vertex on day one. Fine.
The actual story is what Anthropic chose to say out loud alongside the launch: Opus 4.7 is not the most capable model Anthropic has built. Mythos is. Mythos is the zero-day-hunting beast that Anthropic parked inside Project Glasswing last week rather than shipping it to anyone with a credit card. Telling your customers “here’s our flagship, also we have something better but you can’t have it” is either a master class in honest communication or the most successful capability-flex in the history of enterprise software. Probably both.
The read here isn’t the benchmarks — 4.7 will trade blows with GPT-5.4 for a few weeks and then nobody will care. The read is that the frontier labs are now openly operating two tiers of model: the one they sell you, and the one they keep in a vault. Get used to it.
Musk vs. Altman Finally Gets a Trial Date, and the Stakes Are the Company
Jury selection in Musk’s suit against OpenAI begins April 27 in Oakland. The latest filing asks the court to remove Sam Altman and Greg Brockman as officers of OpenAI. Not a settlement. Not a fine. Ouster.
This is the lawsuit that started in 2024 as “you promised me a non-profit and then you didn’t” and has mutated into something closer to a hostile governance takeover attempt through the federal court system. Musk’s lawyers argue that OpenAI’s for-profit pivot was a breach of the original $38 million bargain. OpenAI’s lawyers argue that Musk is a competitor running xAI and is trying to kneecap them via the courts because he can’t via the market. Both of these things can be true at the same time. They probably are.
Whatever the jury does, the trial is going to produce discovery documents that every AI journalist has been salivating over for two years. Expect leaks. Expect subpoenaed Slack threads. Expect Altman’s deposition to be the most-read PDF of the summer.
Q1 Venture Numbers Are In, and They Are Absurd
$242 billion went into AI companies in Q1 2026. That’s 80% of all global venture funding, full stop. Four of the five largest rounds ever recorded closed in the first three months of this year: OpenAI ($122B), Anthropic ($30B), xAI ($20B), Waymo ($16B). Together, those four ate 65% of the quarter.
What this means in practice is that every other startup category — biotech, climate, fintech, SaaS — is now competing for pocket lint. Seed-stage founders pitching non-AI ideas are being told to rewrite their decks. The Magnificent Seven dropped an estimated $78 billion in AI capex in the quarter alone, up 45% year over year. Somebody, somewhere, is going to have to start generating revenue that justifies these numbers. Nobody appears to be in a hurry.
Also Noted
Stalking victim sues OpenAI. Automated safety system flagged the abuser for “mass casualty weapons” activity, a human reviewer restored his account the next day. The complaint is ugly and the discovery phase is going to be uglier.
xAI sues Colorado over its AI law, claiming Grok’s free speech rights. This is the first serious state-vs-frontier-lab preemption fight and it will not be the last.
OpenAI acquired Promptfoo this month, its sixth acquisition of 2026. The open-source eval tool joins Astral inside OpenAI’s dev-tools land grab. Consolidate the pipes, starve the middle layer, call it a moat.