Apple Is Paying Google $1B a Year to Run Siri — The AI Governance Trap
Apple is reportedly paying Google around $1 billion a year to run Siri on a custom Gemini model. Four stories this week, one question underneath each — who controls the AI you run on, and what they let you do with it.
Four stories this week, and one question runs under all of them: who controls the AI you run on, and what do they let you do with it. Here is the week, with the promise and the risk of each side by side.
Apple rents Siri from its biggest rival
At WWDC, Apple revealed that the new Siri AI runs on a custom Google Gemini model, reportedly for around $1 billion a year, inside Apple’s own data centers. This lands weeks after a $250 million settlement over the 2024 Siri features Apple announced and never shipped. The promise is a Siri that finally works, fast. The risk is that the company that built its brand on doing AI its own way now rents its flagship feature from a direct competitor, and the dependency runs both ways.
Anthropic ships Fable 5, gates Mythos 5
Anthropic released Claude Fable 5 to the public this week, held the unrestricted Mythos 5 back over cyber and bio risk, then throttled access to the public model within days of launch. The promise is a lab pricing safety into its release schedule rather than shipping everything it can. The risk is that “Access Throttled” is now a status your production workflow can wake up to, on a model you do not control. The full breakdown is in this week’s deep dive.
Moonshot gives a trillion-parameter coder away
The same week Anthropic gated its strongest model, Moonshot AI put Kimi K2.7-Code on Hugging Face — a one-trillion-parameter coding model with open weights, running up to roughly 12 times cheaper per token. The promise is control: run it inside your own perimeter, your code never leaves the building. The risk is a capable coding agent that answers to no vendor and shows up in no change log. The standalone breakdown is here.
Europe writes the labeling rules early
On June 10 the EU AI Office published its Code of Practice on the transparency of AI-generated content — marking and labeling deepfakes and chatbots ahead of the AI Act rules that land in August. The promise is a concrete standard before the deadline, not after. The risk is that labeling is the floor, not the finish line, and any organization placing AI on the EU market has weeks, not quarters, to be ready.
New rundown every Saturday. Full episodes every Tuesday.