AI Bias &
Governance Policy
The guardrails on how Compass uses AI — approved models, human review requirements, and quality controls.
Effective date: August 3, 2026
This document defines the systemic guardrails and internal audit metrics deployed by Waypoints Digital LLC to manage Artificial Intelligence integrations safely and responsibly within Compass.
1. Approved Model Boundaries and Data Integrity
- Authorized Model Provider: The Anthropic Claude API is the sole authorized Large Language Model (LLM) processing engine integrated into the Compass copywriting tool.
- Input Restrictions: Employees, system admins, and customers are strictly prohibited from inputting proprietary software source code, corporate financials, or guest personally identifiable information (PII) into the AI generation prompt interface.
- Data Training Opt-Out: Our API connection is explicitly configured to opt-out of data sharing for model training purposes. Text prompts sent to the Anthropic API are handled strictly as transactional requests and are not utilized by Anthropic to train future iterations of their core models.
2. Mandatory Human-in-the-Loop Review
- Draft Classification: All social captions, ad copies, and marketing text variations generated via the Anthropic Claude API are classified strictly as unverified drafts.
- No Automated Publishing: The platform does not support or allow fully automated publishing pipelines from the AI text model to live ad networks. All generated variations require manual review and editing by the campground owner or staff before deployment.
- Interface Disclaimer Notice: The generation interface presents the following disclosure to all operators: “All text variations created by this tool are AI-generated drafts. The user maintains sole responsibility to review, edit, and approve all copy for factual accuracy, bias, brand compliance, and legal safety before publishing.”
3. Quality Controls and Review Protocols
Waypoints Digital LLC monitors the performance of the integrated Anthropic Claude API using a risk-based approach. Rather than automated loops, our engineering team reviews system prompt effectiveness and output trends periodically based on operational needs. To facilitate this, Compass features an inline user flagging utility. When a subscriber flags an output as inaccurate or inappropriate, the underlying prompt is isolated for review, allowing our team to fine-tune our system instructions dynamically.