Nadella: We Should Assume All AI Models Are 'Compromised'

It means an organization should not assume a model is fully trustworthy or intact. Instead, it should treat the model as something that could be manipulated, leaked, or misdirected, and surround it with containment and monitoring controls.

A trust-based approach accepts or rejects model outputs as-is. Nadella proposes a transparent system where the model is contained, observed, and leaves verifiable evidence that can be audited later.

The recommendations align with existing regional regulatory directions, such as data protection and technology governance requirements in financial and government sectors. Building containment, audit, and disclosure into AI policies reduces compliance risk.