SECAI Core · phase 6 of 15
Model access
The ability to interact, use, or control an AI model
The Explain card
- Plain English
- Model access is the ability to interact with, use or control an AI model, whether by sending it prompts, calling its API, adjusting its settings, or reaching its weights.
- Example
- A research team's fine-tuned model sits behind an internal endpoint with no authentication, because "it is only on the intranet". A contractor's compromised laptop can now query it freely and extract proprietary behaviour through repeated prompts.
- Why it matters
- Access defines who can abuse, extract or reconfigure a model. Defenders apply the same identity and least-privilege discipline to models as to databases: authenticate callers, scope permissions, and log usage.
- Hook
- Model access is a privilege, not a convenience; grant it like database access.
Where it sits in the deck
Phase 6: Model Quality, Ethics, and Responsible AI Principles
Before deploying a model, understand the quality and ethical dimensions — bias, fairness, explainability, and the human oversight needed to govern outputs responsibly.