Zero-Confidence is an assume-breach posture for systems built, modified, or maintained with AI assistance. It holds that every such artifact carries a nonzero probability of compounding error until Z-sufficiency is demonstrated. Absence of evidence clears nothing.

Under this posture, the burden of proof sits with the party disputing compounding, not the party raising it. Vendor statements, root-cause attributions, and success categories carry zero weight until driven to checkable ground. Exposure signals, outcome signals, and rate changes move confidence; assurances do not.

The posture exists because current instruments cannot see the failure. Qualification regimes treat AI-origin artifacts as identical to human ones. Verification tools erase provenance by design. AI tools used in investigation resist attributing failure to AI, and that resistance is directional. Compounding errors enter upstream, stay latent, cross organizational boundaries unlabeled, and surface downstream after correction is no longer possible.

Zero-Confidence treats AI compounding as the presumptive cause of any loud failure in an AI-exposed system. The presumption stands until provenance is recorded and Z-sufficiency is demonstrated. Every such failure is investigated as a potential system shock event.

The question is not whether to trust the system. It is what has been shown.