Research at Krim
The work under the product.
Krim is a research company first. Everything we ship is that research, in production.
The strands
Hard questions, one body of work.
Each strand is a research problem in its own right. Start anywhere. Each opens onto where it goes deeper.
Validation science
Proving before acting is the harder path. We chose it.
Most AI safety is read after the fact, a log to inspect once something has happened. In regulated work that is too late: a wrongful disclosure cannot be unmade, a misquoted figure cannot be unspoken. So we treat pre-execution validation as its own discipline, checking a proposed action against law, policy, consent and context before it can fire.
The aim isn’t fewer violations. It is to make the worst ones structurally impossible, a research problem before it is a product one.
Pre-execution, not post-audit
The frontier
The open questions are the interesting ones.
How a learned model of an operation and a formal validator compose; how a system improves from its own outcomes without drifting from the rules; how proof stays fast enough to sit in front of every action. This is the frontier we are building on, and the foundations under it are real today.
See the research run, end to end.
The foundations aren’t a roadmap slide. They are a working runtime your teams can watch validate every action, live.