We build our own models, try hard to break them, and release only what we'd trust ourselves. Our first one is called WAVE 1.
Capable AI should be built slowly, in the open, by people who sweat the details.
WAVE 1
A conversational AI. Ask it things; it answers — and when it doesn't know, it tells you instead of guessing. It's in development now, and it ships once it passes our own safety testing. Not before.
Build it. Break it. Only then ship it.
We build our own
Wave AI trains and builds its models in-house — so we understand exactly how they behave, and own the responsibility for it.
We try to break them
Before anyone uses what we make, we red-team it: probing where it fails, where it's wrong, and where it could do harm.
Then we release
A model leaves the lab when it passes our testing — not when a date on a slide says so. We'd rather ship late than ship something we can't stand behind.
Powerful tools deserve patience. We take it.
Build it. Try to break it. Fix what broke. Ship it only if we'd use it ourselves.