Model Choice Is an Information Policy

Choosing a model decides where data goes, what is retained, who administers it and which actions become possible.

The answer

Model selection is presented as a capability decision: accuracy, latency and price. For private work, it is also an information policy. Hosted, regional, local and specialist models create different observers, retention and tool boundaries.

Model selection is presented as a capability decision: accuracy, latency and price.

For private work, it is also an information policy. Hosted, regional, local and specialist models create different observers, retention and tool boundaries.

Route by task and data

A public research question, restricted legal analysis and personal health summary should not automatically use the same model and connector set.

Value object — The Model Routing Policy

- Task class.

- Data classification.

- Approved models.

- Permitted connectors.

- Retention and region.

- Escalation or no-AI route.

Keep choice replaceable

Evaluate factual quality, privacy controls, failure behaviour and portability. Avoid embedding provider names directly into every workflow.

The best model is the model permitted for this information and consequence—not the one winning a generic leaderboard.

Where this breaks

A routing policy can exist on paper while users select models directly from a menu. Sensitive work then follows convenience, speed or brand familiarity.

The operating move

Enforce routing in the interface and connectors. Show users why a task was routed and provide a secure exception process when the classification is wrong.

Default by data class.

Hide prohibited model choices.

Log routing exceptions.

Re-evaluate after provider change.

The test

Submit the same sensitive task through every available interface. None should bypass the approved model, retention and connector policy.

Sources

  1. NIST: AI Risk Management Framework — Generative AI ProfileNIST: AI Risk Management Framework

    Primary authority

  2. Swiss FDPIC: AI and data protectionSwiss FDPIC: AI and data protection

    Primary authority

  3. NIST SP 800-207: Zero Trust ArchitectureNIST SP 800-207: Zero Trust Architecture

    Primary authority

Ross BelhommePartner, Svperior / Legal

Adam J. De Collibus

Adam co-founded Svperior and leads systems engineering from requirements through implementation. His work connects architecture, implementation, deployment, and operating discipline across complex environments where failure must be anticipated and technical capability must remain dependable under pressure.

Systems engineering / Technical architecture / Production operations / Operating resilience

Need to apply this to a specific situation?

Send us the initial context. If the matter fits, we will respond directly.

Send private inquiry
Model Choice Is an Information Policy \