AI Principles, Accountability and Acceptable Use

This page sets out how we build, deploy, and hand over AI systems for the businesses we work with, who is accountable, what we will not build, and how anyone can raise a concern.

It is a trust artefact, written for the directors, owners, and operators who carry the risk of the systems we install. It is not a marketing page. If a claim here does not hold up to a direct question, we have written it wrong and want to hear about it.

1. Our principles

These are the principles that sit behind every engagement. They show up in our scoping conversations, our service agreements, and our handover packs.

• The client owns the system. Every tool, credential, account, dataset, and workflow we build sits under the client's own infrastructure. If Roovolt stops tomorrow, nothing breaks and nothing walks out the door.
• Human oversight is the default. No workflow sends externally, commits funds, or takes an irreversible action without a person in the loop. We design for meaningful human control, not for autonomy.
• Narrow scope beats broad capability. We build the specific system the business needs. We do not deploy unbounded autonomous agents in production, and we do not recommend them.
• We hand over what we build. Every engagement ends with documentation, training, and a 30-day stabilisation period with the client team, so the people who have to run the system after us actually can.
• We tell clients where the data goes. Every engagement ships with a written Security Considerations section covering the specific tools deployed, their data residency, their retention settings, and their known limitations.
• We are honest about what we are not. We are a small Australian consultancy, not a certified AI assurance provider. Where a framework calls for a formal artefact we do not yet produce, we say so and point to how we close the gap.

2. Accountability

Douglas Reyes Corrin is the accountable person on the Roovolt side for every AI system we build, deploy, or hand over. He is the point of contact for Roovolt clients, and for any director or accountable officer reviewing a Roovolt engagement, on any question, concern, or incident relating to our work.

Contact:

• Email: doug@roovolt.com.au
• Phone: 0468 031 661 (Roovolt AI Assistant line, routes to Doug)

For every active engagement, a named escalation path on the client side is agreed during scoping, so that issues have a clear route on both sides.

3. Human oversight, by default

The single strongest claim we can make about our work is that meaningful human oversight is the operating default, not an optional extra. This is consistent with Guardrail 5 of the Commonwealth Voluntary AI Safety Standard (2024).

In practical terms:

• A human reviews AI output before it reaches a customer, supplier, or other external party.
• A human authorises any action that moves money, commits the business, or cannot be reversed.
• A human can override, pause, or switch off any system we have built at any time, using credentials the client owns.
• Where an AI system produces a decision that affects an individual, that decision can be reviewed and contested through a named human reviewer.

4. Acceptable use

What we will build

• Scoped workflow automation that removes repetitive manual work from specific roles.
• AI-assisted drafting, summarisation, and classification, with a human review step before anything leaves the business.
• Voice and chat intake systems that triage and route, with clear disclosure to the caller or user that AI is being used.
• Internal research and analysis tools that a staff member operates directly.
• Systems that help a team adopt AI, with training, documentation, and measurable KPIs defined up front.

What we will not build

• Models trained or fine-tuned on client data for our own use, or for onward use by other clients. Client data stays with the client.
• Unbounded autonomous agents running in production without a documented human-review step.
• Workflows that send externally, commit funds, or take irreversible actions without a human in the loop.
• Systems that produce decisions about individuals (employment, credit, eligibility, benefits) without a named human reviewer and a contest path.
• Automated data collection that breaches a platform's terms of service.
• Systems designed to deceive end users about whether they are interacting with a person or an AI.
• Engagements for clients whose purpose is to use AI to circumvent Australian law, including the Privacy Act 1988, the Do Not Call Register Act, or sector-specific obligations.

If a proposed build sits in a grey area, we say so in writing during scoping. If we cannot find a way to meet the principle above and deliver the business outcome, we decline the work.

5. How to raise a concern

If you are a Roovolt client, or a director or accountable officer reviewing a Roovolt engagement: email doug@roovolt.com.au or call 0468 031 661. Include the engagement reference and a description of what happened. We respond as quickly as we can.

If you are an end user, a customer of a Roovolt client, or another third party interacting with a system we have built: the contest and complaint path sits with the client that owns and operates the system. The client is the accountable party for how the system is used in their business. Your first contact should be their nominated accountable officer within that business. This is how the frameworks above intend contestability to work: with the deployer of the system, not with the supplier that built it.

If your concern relates to a live incident (harmful output, data exposure, unauthorised action) and the client is not responding, you can contact us directly at the details above. We will prioritise it and move to notify the client's accountable officer without delay.

If you are not satisfied with the response and your concern relates to personal information, the Office of the Australian Information Commissioner (oaic.gov.au) is the relevant external avenue.

6. Frameworks that inform this posture

Our delivery approach is informed by, and consistent with the core guardrails of, the following Australian frameworks:

• AICD, A Director's Guide to AI Governance (2024). The 8 Elements of Safe and Responsible AI. We reference this Guide when scoping engagements and structuring handover documentation, and we are progressively formalising our practice against the elements we currently satisfy only in part.
• Commonwealth Voluntary AI Safety Standard (2024). The 10 Guardrails published by the Department of Industry, Science and Resources. Our human-oversight default and client-ownership model directly implement Guardrail 5 (meaningful human oversight). Formal documentation against all 10 guardrails is in progress and tracked internally.
• National AI Centre, Guidance for AI Adoption (October 2025). The evolution of the 10 guardrails into 6 essential practices for organisations adopting AI. We track this guidance alongside the standard above.
• Australia's 8 AI Ethics Principles (DISR). The principles our acceptable-use rules are anchored in.

We do not claim full alignment with any of the above. We claim our operating model is consistent with their core guardrails and that we are honest about the gaps. Directors who want to see how we assess and monitor AI systems internally can email hello@roovolt.com.au and we will share the templates we currently use.

7. Version history

This page will be updated as our practice formalises against the frameworks above. Version and date are at the top of the page. Material changes are noted here.

• v1.0, 14 April 2026. Initial publication.