The AI talent raid that won't slow down

Two more senior Gemini researchers have left Google for Anthropic, making four high-profile Google departures in six days. The model wars have quietly turned into poaching wars, and right now the talent is flowing one direction.

This matters more than it might sound. Frontier AI is built by a small number of people who actually know how to train and align these systems. When four of them leave one lab for a rival inside a week, it isn't just a hiring stat, it's a signal about where the most ambitious researchers think the interesting work is happening, and about how brutal the competition for that handful of people has become. Pay packages for top AI researchers now run into the tens of millions.

The other half of the story

On the same day, Anthropic formally accused Alibaba of running 28.8 million fraudulent exchanges against Claude. The detail is murky and Alibaba will have its own version, but the accusation points at a growing problem across the industry: rivals and bad actors probing each other's models at scale, whether to harvest outputs, find weaknesses, or train competitors on the results.

Put the two together and you get a snapshot of where the AI business is in mid-2026. The technology is maturing, but the competition around it is getting sharper and less friendly. Labs are fighting over a tiny pool of researchers and accusing each other of misusing their systems.

Why it matters

For everyone outside the labs, this is the weather behind the products you use. The pace of model improvements, the price you pay, and which assistant ends up best a year from now all trace back to who hired whom this month. Talent concentration is the real moat in AI, and right now it's moving.