A field map of the verification problem space — what exists, what's missing, and where the open problems live.
Verification for AI is a young field, and it is not spread evenly. Some corners are crowded with research; others are nearly empty. This map lays the work out along two axes: the kind of verification down the side, and the who across the top. The darker a square, the more activity sits there today.
Tap any square to see who is working there and how the pieces connect. Tap a row or column label to read what it means. Reading the heat is the point: notice which corner is dark, and which is blank.
One pattern jumps out fast. The field's center of gravity is the think-tank and nonprofit column, not government or industry. Most of what exists today is analysis and proposal. The hard mechanisms, especially the cryptographic ones that would let a state check a rival without seeing its secrets, sit mostly in academic papers with no public-sector home. The single place where government verification is genuinely muscular is export control, one square in the whole grid.
That unevenness is the opportunity. You are not arriving at a finished field. You are arriving at one with obvious empty squares, at a moment when which ones get filled is still open.
A snapshot, not a census. Efforts are illustrative examples, current to early 2026, and the field is moving quickly. Intensities are judgment calls meant to convey shape, not precise measurement.