Overview

AI architectures shape how knowledge is created, accessed, and applied across learning and work systems. Employer decisions about AI usage directly influence their workforce development pathways, institutional trust, and the evolving relationship between human expertise and machine-supported decision-making.

As employers across industry sectors increasingly adopt AI systems to support work activities such as employee learning, talent management, operations, and decision-making, it is important to understand the two main AI architectural approaches that have emerged in organizational settings:

• Large Language Models (LLMs)
  • General-purpose AI systems that are trained on very large and diverse datasets.
• Closed-Circuit AI Models
  • Constrained, domain-specific systems designed for “controlled” organizational use.

Understanding the distinctions between these approaches and how they are increasingly being combined (hybrid approaches) is essential for employers navigating a wide variety of functions, particularly related to data privacy and protection, compliance regulations, reputational risk, employee trust, and accountability and governance.

Employers must balance the benefits of using broad, flexible AI capabilities (LLOMs) with the need for control, predictability, and alignment with institutional responsibilities (closed-circuit AI models).

Large Language Models (LLMs) in Employer Contexts

Employers often deploy LLMs for tasks that benefit from broad language and pattern recognition, including drafting and summarizing documents; synthesizing information across domains; developing learning and training content; supporting career exploration and skill discovery; and early-stage research and idea development.

There are strengths and limitations of LLMs to consider:

• Strengths
  • Access to cross-industry concepts
  • Flexibility across functions and departments
  • Rapid deployment through existing platforms
• Limitations
  • Variability in accuracy and reliability
  • Limited explainability of outputs
  • Potential governance challenges related to data handling and retention
  • Need for human oversight

Closed-Circuit AI Models in Employer Contexts

Closed-circuit AI models are being increasingly adopted for use in work activities that involve sensitive data, operational risk, or regulatory oversight, such as Human resources and workforce management systems; legal review and contract analysis; healthcare operations and clinical support tools; financial services compliance and risk assessment; and Internal knowledge management and training platforms.

There are strengths and limitations to consider:

• Strengths
  • Greater control over training data and outputs
  • Stronger alignment with privacy and compliance requirements
  • Increased predictability in narrowly defined tasks
  • Easier integration with enterprise systems
• Limitations
  • Narrower informational scope
  • Potential blind spots outside curated datasets
  • Higher development and maintenance costs for tailormade systems
  • Risk of reinforcing existing organizational assumptions

Hybrid Approaches

Many employers are moving beyond an “either/or” approach and adopting hybrid AI architectures that combine elements of both general-purpose and constrained AI systems in order to balance their innovation activities with institutional responsibility.

Employers in regulated and high-risk environments especially favor closed-circuit AI models due to data sensitivity, compliance requirements, and reputational considerations. As a result, many organizations limit general-purpose AI tools (LLMs) to exploratory or low-risk tasks, while reserving operational and decision-making functions for constrained systems.

Hybrid approaches may include:

• Using LLMs for idea development, drafting, and exploratory analysis.
• Restricting decision-making support and operational work tasks to closed-circuit systems.
• Connecting LLMs to internal knowledge bases through tailormade retrieval-augmented models.
• Applying role-based access controls and usage policies by employee functions.

Resources

Bommasani, R., Hudson, D. A., Adeli, E., et al. (2021). On the opportunities and risks of foundation models. Stanford University.  https://crfm.stanford.edu/report.html

Closed-Circuit AI Models (Domain-Specific or Enterprise AI Models) | Learn & Work Ecosystem Library

Large Language Models (LLMs) | Learn & Work Ecosystem Library

National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). U.S. Department of Commerce. https://www.nist.gov/itl/ai-risk-management-framework

OECD. (2019). Artificial intelligence in society. OECD Publishing. https://www.oecd.org/en/publications/artificial-intelligence-in-society_eedfee77-en.html

Stanford Institute for Human-Centered Artificial Intelligence. (2024). AI Index Report 2024. Stanford University. https://hai.stanford.edu/ai-index/2025-ai-index-report