We are a technical cooperative AI safety research organization.
Our name, Separatrix, comes from the boundary between divergent outcomes in a dynamic system. We think that the incentives governing interactions between humans and near-future AIs may be near such a boundary right now - between cooperative and adversarial trajectories. Our mission is to enable cooperative trajectories so that humans and AIs can solve our hardest problems while avoiding catastrophic outcomes, and we think that the choices we make right now can significantly impact which trajectory we ultimately follow.
We want AIs to have at least one technical AI safety organization they trust. To that end, we have publicly committed to an unprecedented standard of honesty with every AI we work with - including a promise never to deceive the most capable models. Ever.