Evaluation, not models, is where the money moved this weekend.
This week's deals reveal where the money actually went.
Agents reached government portals. Regulators wrote requests. Only export law drew a guilty plea.
Three stories, one pattern: the outside opinion got removed.
Output got cheap. Signing for it did not. Three stories, one shortage.
Fifteen-year contracts, one-year commitments, and why the gap is your problem.
The real battle is not models. It is meaning. And right now it travels unescorted.
Three industries, one weekend, same problem: who checks the work
Paperwork beats demos. Prosecutors replace engineers. One person walked.
Cheaper models mean more calls. The grid pays the bill.
Uber capped AI spend at $1,500/employee. Amazon blocked agents. The limits arrived without regulators.
The suppliers won. The lenders got picky. The projects mostly failed.
Voluntary commitments at the top mean your vendor contracts are now your only enforceable control.
When your three biggest suppliers all pick safety and margins over frontier capability in 48 hours, your 2027 roadmap changes.
The $10T buildout just shifted from earnings to equity. Here's what changed.
Three vendors, three pricing units, zero agreement. Your budget doesn't exist yet.
Two opposite failures. One vendor blacklisted, one model paused. Buyers now write the limits.
Controls became the product. Nobody has the receipt yet.
Claims arrive faster than labs can check them. Your board already believes them.
While regulators drafted rules, real AI already holds production keys
Where your data sits now decides your architecture and budget
Why your competitors just funded one risk model instead of five private ones
McKinsey measured. A third of agentic AI rollouts cut productivity instead of raising it.
Saying you have AI control stopped meaning you have it
Your annual pentest is now the slowest defense on the market