This paper proposes Semantic Integrity Constraints (SICs) to address the trustworthiness challenges of AI-augmented data processing systems (DPSs), which integrate large-scale language models (LLMs) into query pipelines to enable powerful semantic operations on structured and unstructured data. SICs generalize existing database integrity constraints into semantic settings, supporting common types of constraints such as grounding, validity, and exclusion, along with reactive and proactive enforcement strategies. We argue that SICs provide a foundation for building trustworthy and auditable AI-augmented data systems. We present a system design for integrating SICs into query planning and runtime execution, and discuss their implementation in an AI-augmented DPS. We also present several design goals, including expressiveness, runtime semantics, integration, performance, and enterprise-scale applicability, and discuss how the proposed framework addresses each goal and remaining research challenges.
