What we’ve done.
In progress
Giving the people accountable for AI a way to watch what it actually does
We hand more decisions to systems acting on our behalf, then piece together from logs what they did — oversight shouldn't come down to guesswork.In progress
Strengthening the reasoning behind a decision by making different intelligences challenge each other
An organization betting everything on a single model inherits its blind spots; reliability begins where different intelligences are made to disagree.In progress
Holding the analysis behind a decision accountable to the outcomes it predicts
A model that sounds confident is cheap; one that keeps score and will be wrong on the record is what a real decision can rest on.In progress
Securing the AI an organization put in front of customers — before an adversary tests it for them
A system already in front of customers is only as trustworthy as the attacks it has survived — and most have survived none.In progress