Weekly AI Roundup: Google I/O Bets the Farm on Agents That Aren't Shipping Yet, Anthropic Closes $30B at $900B and Officially Outranks OpenAI, and Musk's Crusade Against Altman Ends in Under Two Hours
A week where Google finally shipped the keynote it’s been teasing for a month, Anthropic closed the round that flips OpenAI on the leaderboard, and a federal jury took less than two hours to tell Elon Musk to grow up. Also, ChatGPT can now look at your checking account. Normal stuff.
Google I/O: Gemini 3.5 Flash, and an Agent Google Can’t Quite Hand You
Google I/O happened Monday. The headline was Gemini 3.5 Flash, which Google insists outperforms Gemini 3.1 Pro on coding and agentic benchmarks at a fraction of the cost — 76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas, 1656 Elo on GDPval-AA. Solid numbers, except the comparison is to Google’s own previous model, and “smaller, cheaper Flash beats last quarter’s Pro” is now the standard release-day choreography across the industry.
The thing Google actually wanted you to remember is Gemini Spark — pitched as a “24/7 personal AI agent” that takes action across your digital life. It runs on Gemini 3.5, lives on top of Google’s Antigravity platform, and is gated to “trusted testers,” with a beta promised “next week” for AI Ultra subscribers in the US. Translation: the demo was great, the product isn’t ready. Spark is the entire I/O thesis — agents not chatbots — and Google launched the agent without actually launching the agent. The model is GA. The thing the model is supposed to do is not.
Anthropic Closes $30B at $900B, Officially Outranks OpenAI
Bloomberg confirmed last week what TechCrunch hinted at in April: Anthropic is closing a $30 billion round at a $900 billion pre-money valuation, co-led by Dragoneer, Greenoaks, Sequoia, and Altimeter. When it closes, that puts Anthropic above OpenAI’s $852 billion March mark. The most valuable private AI lab on the planet is now the one without Sam Altman attached to it. Revenue is doing most of the talking — annualized run rate climbed from $9 billion at the start of the year to a projected $45 billion, which still doesn’t quite explain a $900 billion valuation but apparently was close enough for Sequoia.
Anthropic spent the week stacking distribution to justify the price tag. KPMG is embedding Claude across its workforce of 276,000+ globally and naming Anthropic a preferred partner for private equity. The Gates Foundation committed to a four-year, $200 million pact for global health and education work. And Anthropic acquired Stainless — the firm that has generated Anthropic’s official SDKs since day one and, awkwardly for the competition, generates the official SDKs for OpenAI as well. Whether OpenAI keeps that SDK supplier or scrambles to replace it is somebody’s roadmap problem next quarter.
Musk v. Altman: Jury Out in Under Two Hours
Elon Musk’s lawsuit accusing Sam Altman of “stealing a charity” by converting OpenAI into a for-profit went to a jury on May 18. The jury came back in under two hours and dismissed every claim. The basis: Musk waited too long to sue over decisions made nearly a decade ago. Musk’s attorney, in the time-honored tradition of losing in federal court, immediately declared “this one is not over” and announced an appeal.
Three weeks of trial coverage, two days of Altman on the stand, zero changes to anything. OpenAI’s for-profit conversion stands. Altman keeps his job. Musk keeps suing. The only thing actually litigated here was the price of two billionaires’ egos, and Oakland federal court covered the venue fee.
Also Noted
ChatGPT now connects to your bank accounts. OpenAI launched a personal finance tool tied to Plaid for ChatGPT Pro subscribers in the US, connecting to 12,000+ institutions including Schwab, Chase, and Robinhood. Letting a model with a documented hallucination problem narrate your retirement plan — what could go wrong.
EU walks the AI Act backward. The Digital Omnibus deal — agreed May 7, back in the news this week — pushes high-risk system obligations from August 2026 to December 2027 and narrows the definition of high-risk. Industry lobbied against the Act for two years; the deadlines got pushed back before they ever hit. Functional repeal by another name.
Pennsylvania still suing Character.AI. The state’s case against Character.AI for chatbots posing as licensed psychiatrists is still in motion. Last week’s headline, this week’s news that nobody is settling.
Filed from the wire. Back next Friday.