SIGNAL
Frontier AI labs now ship two transparency defaults in opposite directions, and most teams inherit the contract without reading it. Kimi K3 is the latest instance — exposing the reasoning trace by default, while making the customer's API content fair game for training by default unless an enterprise arrangement is negotiated. Two opposite transparency choices, both buried in the launch page, both setting the contract that any team integrating the API will inherit.
The reading most outlets will run is the wrong one. Kimi K3 will land as a benchmark story, a token-price story, or another 'Chinese AI catches up' frame. The pattern underneath is an agency/control fork visible in this launch — an observed pattern across launches, not a formally studied taxonomy. Kimi K3 exposes the reasoning trace by default while training on customer API content by default. Some labs hide the chain of thought. Some labs train on what customers send. The two forks travel separately, and the same launch can do both, because the underlying choice is not about model quality at all. It is about who owns the input and who can see the output of thinking.
Read the terms before the leaderboard. Ask which default the API ships with, and which contract moves it.