The rapid evolution of modern generative AI architectures (option B) is the largest barrier to explainability.

Complex deep learning models like LLMs, diffusion models, and transformer-based architectures involve millions or billions of parameters, making it extremely challenging to determine precisely how outputs are produced.

AAIA notes that explainability challenges arise because:

Model structures are highly complex

Parameter interactions are nonlinear

Internal representations are not human-interpretable

Continuous updates make documentation outdated

Training data and latent representations create opaque reasoning chains

Bias (A) affects fairness, not explainability.

Stakeholder alignment (C) is a governance issue.

Lack of staff experience (D) is a training problem, not a structural barrier.

The inherent technical complexity and speed of model evolution are the primary obstacles.