AI Architecture

Design and structure of AI systems affecting compliance and capability

Core AIModelsTechnicalLegal
Back to Glossary
Updated 18 October 2025

Definition

AI architecture refers to the overall design and structure of an artificial intelligence system. This includes how its algorithms, neural networks, data processing pipelines, and hardware components are organized to perform tasks such as classification, prediction, or decision-making.

Legal Context / Relevance

AI architecture plays a key role in determining whether a system is subject to legal obligations under emerging AI regulations. For example:

  • Under the EU AI Act, architectural features (e.g. deep learning, real-time data inputs) can affect whether a system is classified as high-risk.
  • In algorithmic accountability laws (e.g. NYC Local Law 144), transparency and auditability of system architecture are compliance-critical.
  • Some frontier AI governance frameworks require architectural disclosures, including model size, training flow, and safety mechanisms.

Examples in Practice

  • A CNN like ResNet is an architecture optimized for image classification.
  • Transformer architectures power language models such as GPT.
  • A multi-component AI pipeline with human-in-the-loop review is commonly used in regulated sectors.

Applications & Use Cases

  • Compliance evaluation
  • High-risk system classification
  • AI lifecycle management

Risks & Considerations

  • Black-box system opacity
  • Hidden architectural dependencies
  • Misclassification under law