AI can prepare the work. Evidence and human authority still control it.

Where AI may be used

• Extract relevant passages from declared sources
• Compare language or patterns across independent records
• Prepare summaries, hypotheses, and draft recommendations for review
• Assemble a review output from findings that already passed evidence checks

What AI output cannot establish by itself

A model response is not evidence of company performance, customer intent, legal compliance, or factual impact. It cannot create a quote, number, client result, or certainty statement that is absent from the source. It also cannot approve outreach, publishing, budget changes, or writes to a live system.

Customer data boundaries

Northstar Stack does not use customer data to train a general-purpose Northstar Stack model. Customer material stays within the declared engagement purpose and is not used as public proof or another customer's evidence.

Provider handling can vary by product, contract, and configuration. The relevant provider and data-use terms are reviewed before customer material is transferred.

AI provider disclosure

Customer agreements identify every AI or processing provider that may receive customer data, its purpose, the data category, and applicable retention settings. The public Subprocessors page covers standing providers.

Human review and evidence gates

Before a client-facing output is delivered, quotations are checked against the source, estimates are labeled, findings retain evidence links, and unsupported impact claims are rejected. A named person owns the recommendation and the consequential decision that follows.

See the Documentation for evidence labels and the Subprocessors page for provider disclosure rules.