Elon Musk’s Worst Enemy in Court Is Elon Musk
Five hours of Elon Musk on the stand and the headline writes itself: a man worth hundreds of billions of dollars, surrounded by lawyers, somehow managed to make Sam Altman look sympathetic. OpenAI’s attorneys didn’t need much — just Musk’s own words, his own tweets, his own emails — and the case started looking less like a principled stand against AI commercialization and more like a billionaire scorned. The courtroom drama here is genuinely must-follow, not just for the gossip, but because the outcome could determine whether OpenAI gets to go public at all.
Sources: Anthropic Could Raise a New $50B Round at a Valuation of $900B
Let that number sink in: nine hundred billion dollars — for a company that, by most accounts, is still burning cash at an impressive clip. Anthropic has received multiple pre-emptive offers in the $850B–$900B range, which means investors are basically paying a premium just to get in the room. Whether this is rational exuberance or something more historically recognizable, I’ll leave as an exercise for the reader.
Anthropic Launches Cowork, a Claude Desktop Agent That Works in Your Files — No Coding Required
While investors are apparently throwing $900B at Anthropic’s door, the company is doing the less glamorous work of actually shipping product. Cowork takes the magic of Claude Code — which has been quietly impressing developers — and brings it to regular humans who just need help wrangling documents without learning to prompt-engineer their way to salvation. Notably, Anthropic’s team reportedly built the entire feature in about ten days using Claude Code itself, which is either a testament to AI-assisted development or the world’s most effective product demo.
Google Search Queries Hit an ‘All Time High’ Last Quarter
Remember when everyone said AI would destroy Google Search? Sundar Pichai would like a word. Search hit all-time-high query volume in Q1 2026, and Pichai credited AI experiences — AI Overviews, Gemini integrations — with driving that growth, not cannibalizing it. I remain curious how long that story holds as AI answers get more complete, but for now, Google is laughing all the way to the data center.
Emergency First Responders Say Waymos Are Getting Worse
“I believe the technology was deployed too quickly in too vast amounts, with hundreds of vehicles, when it wasn’t really ready” — and that’s a police official telling federal regulators, not some tech skeptic with a blog. Waymo has been the gold standard of autonomous vehicles for years, so hearing that first responders are finding interactions with these cars increasingly problematic is worth paying close attention to. The gap between demo reel and real-world deployment has a habit of biting hard.
Yet Another Experiment Proves It’s Too Damn Simple to Poison Large Language Models
A security engineer spent $12 on a domain, edited one Wikipedia article, and convinced multiple AI chatbots that he was the reigning world champion of a German card game — a title that doesn’t even exist. This is less a security research paper and more a controlled demonstration of what happens when you build confident-sounding oracles on top of unverified web scrapes. The gap between “sounds authoritative” and “is accurate” remains AI’s most underappreciated liability.
Microsoft Lifts 2026 AI Spend by $25 Billion to Cover Component Price Rises
Microsoft is now looking at $190 billion in capital expenditure for 2026 — $25 billion more than planned, thanks to rising hardware costs. To put that in perspective, that’s more than the GDP of a mid-sized country, spent in a single year, by a single company, mostly to make sure its AI products don’t run out of compute. The AI infrastructure arms race has moved well past “expensive” into territory that requires entirely new vocabulary.
OpenAI: Where the Goblins Came From
This one’s more technical than dramatic, but genuinely interesting: OpenAI published a post-mortem on why GPT-5 started exhibiting what they delicately call “personality-driven quirks” — internally nicknamed goblin outputs. It’s a rare, candid look at how unexpected behaviors propagate through large models and then get traced back and patched, and it’s the kind of transparency the industry needs more of. Also, “goblin mode” just got an official origin story.
Bottom Line
From a $900B AI startup valuation to a $12 Wikipedia edit breaking chatbots, today is a reminder that AI in 2026 operates simultaneously at the scale of human ambition and the fragility of human error.