Principles / The Trust Ratchet
§ Principle 22 of 23
The Trust Ratchet
Trust in AI systems is hard to build and easy to destroy. Each success adds incrementally; each failure subtracts dramatically.
The Trust Ratchet describes the asymmetric economics of trust. A hundred successful interactions might build moderate trust. One undetected Execution Hallucination can destroy it. This asymmetry means that trust-preserving design isn't about maximizing successful outputs — it's about minimizing undetected failures. A system that succeeds 99% of the time but hides its 1% failure rate will eventually lose user trust catastrophically. A system that succeeds 95% of the time but clearly flags and explains its 5% failure rate maintains trust because the failures are visible and manageable.
Why it matters
The Trust Ratchet means that transparency about failures is more important for trust than consistency of success. Systems should be designed to fail gracefully and visibly rather than to hide failures and hope they go unnoticed. When users discover hidden failures — and they always do — the trust cost is far greater than if the failure had been acknowledged upfront.
In practice
Design your system to surface its failures clearly. "I'm not confident about this output — here's why" builds more trust than "Here's your output" followed by a silent failure. Users forgive acknowledged limitations; they don't forgive hidden ones.