Stage 1: Free safety marketplace

AI alignment can't happen in the dark.

Every AI lab is working on safety. But they're all doing it alone—building private datasets, running isolated benchmarks, keeping their progress behind closed doors. This is a shared human problem, and we're treating it like a competitive advantage.

Data Canon is fixing this. We're building the first transparent marketplace for safety training data—starting with zero fees, proven lift on real benchmarks, and a public commitment to making every safety dataset discoverable.

Set the standard

We're bringing together the industry's leading safety and alignment benchmarks into one place. Shared standards, consistent measurement, so every lab can compare progress on the same foundation.

Prove what works

No more marketing claims. We train open source models on real datasets and publish the measured lift on safety benchmarks. We also test labs' models against these datasets. Data vendors get independent validation. Labs get confidence before they buy.

Zero fees, full access

Safety datasets belong on a marketplace that doesn't tax progress. We host, we don't take transaction fees. If a dataset improves alignment, it should be easy to find and easy to use.

Make it public

We're circulating a commitment to safety data transparency—asking every lab and vendor to bring their training data into the light. You don't have to make it free. You do have to make it discoverable.

Safety benchmarks

A public view of the benchmarks and model scores that define safety progress.

Dataset marketplace

A curated catalog of safety datasets, access terms, and evidence of benchmark lift.

Transparency commitment

A contract for labs and vendors willing to bring safety training data into the light.