Getting Copilot-ready as a NDIS provider: a 6-step plan for safe rollout without exposing participant data

Posted on by

TL;DR

  • Copilot doesn't leak data. But it does reflect whatever's already in your SharePoint. If permissions are messy, Copilot will happily surface participant records to people who shouldn't see them.
  • Your NDIS Code of Conduct obligations don't pause for AI. Principle 2 ("respect the privacy of people with disability") and the Privacy Act 1988 still apply. Copilot is just a faster way to break them if your foundations are weak.
  • The fix isn't fancy. It's a content audit, tighter permissions, a clean-out of legacy files, sensitivity labels for participant data, staff training (especially around "shadow AI"), and a small pilot before you scale.
  • The NDIS Commission is watching. Their February 2026 AI Transparency Statement makes it clear: providers using AI need governance, transparency, and human oversight built in from day one.
  • Want the shortcut? Take our free Is your SharePoint ready for Copilot? survey. You'll get a prioritised list of what to fix first.

Why this matters right now

If you're an NDIS provider, you've probably had at least one conversation about Copilot in the last six months. Maybe from your CEO who saw it demoed at a conference, maybe from a team leader who's tired of writing case notes at 9 pm, maybe from a board member asking, "are we doing AI yet?"

It's a fair question. Copilot can genuinely save your team hours every week: drafting reports, summarising long email threads, pulling together meeting notes, and finding that policy document nobody can remember the name of. For a NDIS workforce stretched thin by paperwork, that's not a nice-to-have. That's giving frontline support workers their evenings back.

But here's the thing: Copilot is only as safe as your SharePoint.

Microsoft has locked down Copilot itself. Your prompts and your data aren't used to train the foundation models. Everything runs inside the Microsoft 365 service boundary on Azure OpenAI. Not the public ChatGPT. Copilot respects the permissions your users already have and only surfaces content they could already access through normal Microsoft 365 search.

Read that last sentence again. Content they could already access.

That's the whole game. If your SharePoint has sites where permissions inheritance is broken, libraries shared with "Everyone except external users," or a public Team that quietly contains a folder of participant plans, Copilot will find that content the moment someone asks the right question. It's not a Copilot problem. It's a SharePoint hygiene problem that Copilot makes visible at speed.

The 6-step plan

Step 1: Get the lay of the land. Audit your content and classify the risk

Before you switch anything on, you need to know what's in your SharePoint. Not what you think is in there. What's actually in there?

For most NDIS providers we work with, there's usually:

  • A "Participants" library that started clean in 2019 and has since become a graveyard of duplicate plan versions
  • A general staff site with a folder called "Old stuff, don't delete" (it's always there)
  • Three or four legacy Teams from restructures past, still containing live participant data
  • That one shared drive someone migrated from the old server "temporarily" in 2022

Run a content audit.

Identify where your high-sensitivity content lives. Think participant records, plans, incident reports, complaints, health information, NDIS numbers, financial information.

Microsoft's SharePoint Advanced Management (SAM) tools can help here, and so can a structured manual review of your top 20 most-used sites.

The goal: a simple map of where sensitive participant data lives, who currently has access to it, and what shape it's in.

Step 2: Lock down permissions and stop the oversharing

This is the biggest single thing you can do to make Copilot safe.

The classic SharePoint oversharing patterns we see across the NDIS sector:

  • "Everyone except external users"as the default sharing scope (it sounds restrictive, but it means every staff member, contractor, and volunteer in your tenant)
  • Broken permission inheritancewhere someone broke a folder out for a one-off project and never put it back
  • Public Teamsthat were set up "just for the team to collaborate" and now contain participant-identifiable content
  • Anyone-with-the-linksharing on individual files that have lingered for two years

Walk through each high-risk site from Step 1. Tighten the default sharing scope. Switch high-sensitivity Teams and sites to private. Review and remove anonymous links. Use SAM's Data Access Governance reports to spot oversharing patterns at scale.

Think of it like this: your SharePoint is a building, and Copilot is a very efficient new staff member who can read every document in every room they're allowed to enter. Before you give them a swipe card, walk the building and make sure the swipe card opens doors to the right rooms.

Step 3: Clean out stale, duplicate, and legacy participant content

Old data is risky data. A participant's plan from 2021 that's still sitting in an unarchived folder is a privacy obligation you've forgotten you have. And a piece of content Copilot can surface and summarise as if it's current.

This step is unglamorous but high-value:

  • Set up site lifecycle policies so sites get reviewed, archived, or deleted on a schedule
  • Identify and retire duplicate participant files(the four copies of the same plan with slightly different filenames)
  • Move old participant records into a properly governed archive with restricted access
  • Apply retention policies that match your obligations under the NDIS Practice Standards and Privacy Act

Your records-retention obligations don't disappear just because the document is in SharePoint instead of a filing cabinet. But the right retention and lifecycle policies mean your live environment stays clean, and Copilot has less stale, sensitive content to stumble into.

Step 4: Build NDIS-aware sensitivity labels and data loss prevention

Sensitivity labels and Microsoft Purview's data loss prevention (DLP) policies are how you teach Microsoft 365 to recognise your sensitive data. Not just credit card numbers, but the things that matter in a NDIS context.

Set up labels and DLP rules for:

  • NDIS participant numbers(a structured identifier you can pattern-match)
  • Plan and budget documents(use a metadata tag, or train a trainable classifier)
  • Incident reports and behaviour support plans
  • Medicare numbers, health information, dates of birth

Once you have labels in place, you can do real things with them: block them from being shared externally, prevent them from being saved to personal OneDrive, alert your privacy officer when they're being moved around, and, crucially, give Copilot the signals it needs to handle that content appropriately in summaries and outputs.

Step 5: Train your people on shadow AI and the NDIS Code of Conduct

Here's the truth nobody likes: if you don't give your staff a safe AI tool, they'll find an unsafe one.

They're already doing it. Someone in your organisation is pasting case notes into the free version of ChatGPT right now to "tidy them up." That's shadow AI, and it's a far bigger participant-privacy risk than Copilot ever will be.

Public AI tools generally:

  • Store the data you paste in
  • May use it to train future models
  • Often process data overseas
  • Have no contractual obligations to your participants

Your training program for Copilot rollout should cover:

  • What Copilot is and isn't(it's a workplace tool inside your tenant. Not the same as ChatGPT)
  • Why public AI tools are off-limits for participant data
  • How the NDIS Code of Conduct applies to AI use. Particularly Principle 2 (privacy) and Principle 4 (acting with integrity and transparency)
  • Practical "good prompt / bad prompt" examples using your own scenarios
  • The human-in-the-loop rule: Copilot drafts; humans review, sign, and own the output

The NDIS Quality and Safeguards Commission's Artificial Intelligence (AI) transparency statement (published February 2026) makes it clear that providers using AI are expected to maintain governance, transparency, and meaningful human oversight. Your training is how you operationalise that expectation.

Step 6: Pilot small, measure, then scale

This is the lesson we keep coming back to from every successful Copilot rollout we've supported: start small, prove it works, then expand.

For NDIS providers, that means:

  • Pick one team or functionfor a pilot (corporate services, HR, or your quality team are usually safer starting points than direct support staff)
  • Define what success looks likebefore you start (hours saved per week, types of work it's used for, any near-miss privacy moments)
  • Run a 6–8 week pilotwith regular check-ins
  • Capture lessonsabout prompts that worked, content that surfaced unexpectedly, and gaps in your permissions or labels
  • Fix the issues you find. Then expand to the next group

Resist the temptation to flip Copilot on for the whole organisation on day one because the license came with it. The providers we see succeed with Copilot are the ones who treat the rollout like any other clinical or operational change: governed, measured, and people-first.

A note on the "people-first" bit

We end nearly every conversation about Copilot the same way: technology is the easy part. Getting your foundations right, training your people, and respecting the trust your participants have placed in you. That's the work.

For NDIS providers, the brand promise is participant privacy.

Get the rollout right, and Copilot becomes a tool that gives your staff hours back to spend with the people they support. Get it wrong, and you've put the most sensitive information held by your organisation in faster, more accessible reach. The difference between those two outcomes is six steps and a bit of intent.

Not sure where to start?

If you've read this far and your honest answer is "I genuinely don't know what's in our SharePoint, let alone whether it's Copilot-safe", you're in good company. That's where most of the NDIS providers we talk to start.

We built a free "Is your SharePoint ready for Copilot?" survey for exactly this moment.

In ten minutes you'll get a prioritised list of the specific things to fix in your environment before you switch Copilot on, in language your CEO and your IT lead can both read.

It's the same readiness check we run at the start of every Copilot engagement. We just made it free, because the alternative, providers rolling out Copilot blind, is worse for everyone.

FAQs

Does Microsoft 365 Copilot send participant data to OpenAI or ChatGPT?

No. Copilot runs on Azure OpenAI services inside the Microsoft 365 service boundary. Your prompts, the content Copilot reads, and the responses it generates aren't used to train OpenAI's foundation models and aren't shared with OpenAI. It's a different product from the public ChatGPT, even though both use related underlying technology.

Can Copilot see participant data that staff don't already have access to?

No. Copilot respects existing Microsoft 365 permissions. It can only surface content that the user asking the question already has access to through normal SharePoint or Microsoft Search. The risk isn't that Copilot bypasses permissions; it's that it makes existing oversharing visible faster.

Does using Copilot breach the NDIS Code of Conduct?

Not inherently. Copilot is a tool, and like any tool, what matters is how you use it. The Code of Conduct applies to your conduct as a provider, including how you protect participant privacy (Principle 2) and act with integrity (Principle 4). If you've got the foundations right (permissions, labels, training, governance), Copilot can support compliance. If you haven't, it can accelerate breaches.

What about the Privacy Act and the Australian Privacy Principles?

NDIS providers are generally APP entities under the Privacy Act 1988. APP 6 (use and disclosure) and APP 11 (security of personal information) are particularly relevant, and APP 8 if your AI processing happens overseas. Microsoft 365 Copilot processes data inside your tenant's geographic region by default, which helps with APP 8, but you still need to assess your own configuration.

How long does getting ready for Copilot typically take?

For a mid-sized NDIS provider with a reasonably tidy environment, expect 6–12 weeks for the core readiness work (audit, permissions, labels, training) before you switch Copilot on broadly. Larger or messier environments take longer, sometimes 4–6 months. The good news is that most of this work is overdue anyway, with or without Copilot.

Do we need to tell participants we're using AI?

The NDIS Commission's AI Transparency Statement (February 2026) signals a clear expectation that providers be transparent about AI use. Even when it's not strictly required, telling participants how you use AI in your operations is good practice, builds trust, and puts you ahead of where the regulator is headed.

Can Copilot help our support workers in the field?

Yes, but be thoughtful. Copilot in Word, Outlook and Teams is genuinely useful for case notes, reports and emails. But for direct participant-facing scenarios (clinical decisions, behaviour support, incident classification), keep the human firmly in the loop. Copilot drafts; your trained staff decide.

Sources

 

And in the NDIS world, "visible at speed" can mean a participant's diagnosis, behaviour support plan, or incident report showing up in a Copilot summary on the wrong person's screen. That's a Code of Conduct breach, an Australian Privacy Principle breach, and, far more importantly, a breach of trust with a person who relied on you to handle their information with care.

The good news? This is fixable. And you don't need to boil the ocean to get there. Here's the plan we walk our NDIS clients through.