Responsible Artificial Intelligence

Design Principles

NOVARIS is designing its future artificial intelligence capabilities around principles of control, transparency, and responsibility.

Permissions Remain the Boundary

An artificial intelligence capability should not have greater access or authority than the authorized context in which it operates.

Responding Is Not the Same as Acting

The system should distinguish between analyzing information, recommending an action, and executing an action.

Operations with meaningful impact should remain subject to the appropriate permissions and controls.

Traceability

When an AI capability participates in an operational action or decision, the system should retain sufficient evidence to understand what occurred and under what authority.

Customer Data Belongs to the Customer

Access to information should follow authorized permissions, context, and purpose.

AI Does Not Automatically Replace Sensitive Human Decisions

Artificial intelligence capabilities are not intended to autonomously make sensitive decisions about people without appropriate safeguards, controls, and review.

Models Can Be Wrong

Artificial intelligence systems can produce incorrect, incomplete, or inaccurate responses. Their design should account for uncertainty, verification, and safe failure mechanisms.

These statements describe the design principles NOVARIS applies to artificial intelligence in its products, including the NOVARIS Go assistant and NOVARIS Pro, which is coming soon.

In NOVARIS Go, the assistant uses Anthropic models (Claude) through Vercel AI Gateway; voice notes are processed with OpenAI, and the assistant’s voice is generated with ElevenLabs. Customer data is not used to train artificial intelligence: we use business services that do not train on the data sent to them. NOVARIS does not publish claims about accuracy, results, or savings that cannot yet be demonstrated.