AI policy.

What this is

This policy explains how Dobbie Engineers uses AI tools on your work, and the choices you have. It covers the AI-specific part of what we do. For how we handle your information generally, including overseas transfer, retention, your rights, and complaints, see our Privacy Statement. The two are meant to be read together.

It also sits alongside our internal AI use procedure, which identifies the tools we have approved and sets the rules our team follows.

How we use AI

AI helps us do good work faster. The tools we use sit in three groups:

• General-purpose large language models like Claude (Anthropic). We use these to draft, summarise, edit, and analyse documents, and to support technical research and coding.
• Meeting transcription tools like Fireflies and the meeting summary features in Microsoft Teams. We record and transcribe meetings only with everyone's consent.
• Workplace AI features in Microsoft 365 (Word, Outlook, Excel, PowerPoint, Teams Copilot). These help with drafting, summarising, and finding information across documents we already have access to.

We use commercial or enterprise tiers of every approved tool. These tiers carry a contractual commitment from the provider not to use our inputs to train public AI models.

An engineer always reviews AI output before it is used in your work. AI does not replace our engineering judgement, our checking, or our sign-off. We do not sign off engineering deliverables on the basis of AI output alone, and we do not rely on AI as the sole basis for any safety-critical decision.

What goes into AI tools

This policy applies to all our engagements and is published here so you can read it at any time. We may put project information through our approved AI tools, including tools run by providers based overseas, for the sole purpose of delivering the agreed engineering services.

Where your contract with us, or your organisation’s own rules, require your approval first or set tighter limits, we follow those. You can opt out, restrict that use, or withdraw at any time. See Your choices below.

For most engagements that information includes:

• Project documents you send us (drawings, specifications, reports)
• Email correspondence about the project
• Meeting recordings and transcripts where consent was given
• Notes, calculations, and working files we produce
• Publicly available information we use as reference

We use the least information needed for the task, and where it makes sense we anonymise, generalise, or summarise it first. For higher risk work, such as sensitive personal information, health and safety incidents, legal advice, or confidential commercial detail, we apply extra safeguards under our internal procedure and may decide not to use AI at all.

What we won't do

• We will not train public AI models on your information or our work for you.
• We will not use your information for marketing, advertising, or any purpose outside the engagement you have agreed.
• We will not record or transcribe a meeting without your consent. Where we use a meeting assistant such as Fireflies, it joins as a visible participant that the meeting host must admit. No AI assistant sends emails on our behalf without our direct review.
• We will not put your information into a personal AI account, a free public version of an AI tool, an unapproved browser extension, or any AI plugin Dobbie has not approved.
• We will not put credentials, passwords, access tokens, or security-sensitive client system details into any AI tool.
• We do not rely on AI for legal advice. Regulatory and compliance conclusions in our work are reached, checked and signed off by our engineers, never by AI alone.

Your choices

You have a clear set of choices about AI use on your work:

• Opt out. Tell us, and we will deliver your work without using AI tools. Many engineering tasks can be done the traditional way without slowing the programme.
• Restrict use. Tell us at the start of the engagement if some material must not go through AI tools, even if other material can. We will set that boundary in writing.
• Choose the tools. If your organisation approves only particular AI models for its information, tell us which ones. We will use only those on your work.
• Ask what we used. If you want to know whether AI was used on a particular document, calculation, or meeting summary, ask. We will answer honestly and accurately.

If your request means we cannot safely complete the engineering work, we will explain why before we decide what to do.

Privacy and your information

Most of the AI tools we use process information on servers outside New Zealand. We check each tool's privacy terms, security controls, retention settings, and no-training commitments before we approve it. How we handle that information generally, including overseas transfer under the Privacy Act 2020, how we protect it, how long we keep it, your rights of access and correction, and how to raise a concern, is set out in our Privacy Statement.

We keep the information you send us for as long as we need it for the engagement and for our professional, legal, and insurance obligations, which for engineering work includes the period we remain liable for our designs. We take reasonable steps to return your information when you ask. Because we run routine backups across our files, email, and document systems, copies may remain in our secure backup systems for a time and cannot always be fully removed.

We review this policy at least once a year, and whenever we add, change, or remove an AI tool that materially affects how we use your information. The current revision is shown at the top.