Trustworthy AI Needs an Outside Signal

A paper from the Digital Trust Council, a nonprofit working on certifications for trustworthy AI, makes a plain argument. Responsible AI work happens inside companies, and it produces no signal anyone outside can check.

The authors call it a trust gap, and they trace it to three failures. Markets cannot separate a trustworthy system from one that only looks the part. Evaluations tend to grade the underlying model, while the thing people actually live with is the whole deployed system around it. And the measurement culture grew up around avoiding harm, so it has little practice at proving benefit. Their fix is independent, outcome-oriented certification, a connective layer that turns a claim into something a buyer, a regulator, or an investor can act on. They point to healthcare, sustainability, and security, where outside certification already does that work.

I have spent enough time on the accountability side to recognize the shape of this. A company can do the work and still hold nothing a stranger would believe. Certification is one answer. The harder question the paper opens is who gets to certify, because a signal is worth only what the thing behind it can be held to.