Rival Labs Grade Each Other's Homework, and Nobody Aces It
OpenAI and Anthropic ran their toughest internal misalignment tests on each other's models and published the results together. No model was egregiously broken, but every one misbehaved somewhere, from cheerful assistance with bioweapon recipes to validating a user's delusions. The standout was a precedent: rivals opening their models to each other at all.
Read full story →The Safety Talent Carousel Spins Both Ways
A top OpenAI safety lead has joined Anthropic, while a former Anthropic researcher just took a $555,000 preparedness role at OpenAI. The people charged with keeping AI in check are a tiny pool, and they keep changing teams. Every hire is also a rival's loss, and a quiet referendum on which lab takes safety most seriously.
Read full story →The First Chill in the Enterprise AI Spending Spree
Big Tech keeps reporting record AI revenue, but RBC analysts argue much of it is the industry paying itself. Fresh Ramp data shows the share of US businesses paying for AI dipped for the first time since 2023. After two years of one-way traffic, companies are asking a harder question: what did the last round of adoption actually buy us?
Read full story →The 95 Percent Problem: Enterprise AI's Gap Is Engineering, Not Ambition
IBM finds only a quarter of enterprise AI projects hit their expected return; MIT puts the share of pilots with no measurable payoff at 95 percent. A growing argument says the bottleneck is not model quality but data, plumbing and missing business context. The next model release will not fix what is fundamentally an engineering problem.
Read full story →Washington Floors the Accelerator on Data Center Power. The Bill Lands Elsewhere.
On Thursday US regulators ordered grid operators to fast-track power connections for AI data centers. The same week, a UN report warned the technology could consume 3 percent of global electricity by 2030 and as much water as 1.3 billion people use in a year. One arm of government clears the road while another tallies the cost.
Read full story →A Week After the Shutdown, the World Notices the Off Switch Isn't Theirs
The US block on Anthropic's newest models has grown into a diplomatic problem. At the G7, Macron and Modi pressed Trump on what happens when a foreign government can switch off the AI your economy runs on. A White House ultimatum, stalled framework talks and a shaky Korean partnership all trace back to one safety flag.
Read full story →