AI Ethics Policy
Last updated: 1 March 2026 · This policy governs all AI systems developed and deployed by SOTAStack AI
At SOTAStack AI, we believe AI should make people more capable, not replace human judgement on consequential decisions. This policy sets out our commitments to responsible AI development and deployment — both for our own products and for the systems we build for clients.
1. Our AI Principles
Our AI systems are built and operated according to five core principles:
AI systems should create genuine value for the people and organisations they serve, not just optimise metrics.
Users should know when they're interacting with an AI system. We don't build systems designed to deceive.
Systems are tested before deployment, monitored in production, and designed to fail safely when they encounter edge cases.
Personal data is used only as necessary for the stated purpose. We prefer self-hosted, Australian infrastructure where possible.
Consequential decisions — medical, legal, financial — always have human oversight. AI assists; humans decide.
2. Transparency
AI disclosure: All SOTAStack AI products that interact with end users (Harbor, VotreV, Smokey) disclose their AI nature. Users are not deceived into believing they are talking to a human unless they have explicitly chosen to configure a product persona without disclosure — and we advise against this practice.
Explainability: Where technically feasible, our systems provide source citations for answers (particularly in RAG-based systems like Holdpoint AI). We actively avoid systems that make confident claims without traceable reasoning.
Model disclosure: We disclose the foundational AI models powering our systems to clients upon request. We do not misrepresent third-party models as proprietary technology.
3. Safety & Reliability
Before deploying any AI system to production, we conduct structured testing across a range of scenarios including adversarial inputs, edge cases, and failure modes. We aim to understand how a system fails before it fails in front of a customer.
Our systems include guardrails to prevent harmful outputs. For systems deployed in regulated industries (health, construction), additional domain-specific safety constraints are applied.
We maintain monitoring and alerting on production AI systems. Significant errors or safety events are reviewed within 24 hours and remediated with priority. Clients are notified of safety-relevant incidents promptly.
4. Privacy & Data
We do not use client data to train or fine-tune AI models without explicit written consent. Data provided for one client's RAG system is never used in another client's system.
We prefer self-hosted, Australian infrastructure to maximise data sovereignty. Where third-party AI APIs (such as OpenAI or Anthropic) are used, we configure them to opt out of training data usage where available, and we disclose this to clients.
Conversation logs and AI-processed data are retained only as long as necessary for service delivery and quality assurance. Clients can request deletion of their data at any time.
5. Human Oversight
We believe in human-in-the-loop design for consequential decisions. Our AI systems are designed to assist and augment human decision-making — not to make final determinations on matters with significant impact on individuals.
All our conversational AI products (Harbor, VotreV, Smokey) include human handoff capabilities. Users can escalate to a human agent at any point. Urgent or sensitive situations are automatically flagged for human review.
For health-sector deployments (DentaVoice), emergency triage protocols route urgent cases to human staff immediately. The AI system does not make clinical recommendations.
6. Accountability
SOTAStack AI takes responsibility for the AI systems we build. We do not build systems intended to deceive, discriminate, surveil individuals without consent, or cause harm.
We review our AI ethics commitments periodically and update them as the technology landscape and regulatory environment evolves. This policy aligns with Australia's Voluntary AI Safety Standard and the principles of the National AI Framework.
If you have concerns about the ethical use of any SOTAStack AI system, please contact us at ethics@sotastack.com.au. We take all concerns seriously and will investigate and respond within 5 business days.