The DTA just told agencies to run AI at scale. Here is what vendors will need to show
The short answer: On 4 August 2026, the Digital Transformation Agency told the public sector that the AI conversation has moved from pilots to whole-of-government scale, and told industry to design for transparency, interoperability, cost visibility and changing assurance needs from the start. If you sell into government, or use AI anywhere in your delivery model or your bid process, expect buyers to start asking how that AI is governed, not just whether you use it.
The speech was delivered by Lucy Poole, Deputy CEO, Strategy, Planning and Performance Division at the Digital Transformation Agency, at the 2026 Tech in Gov event. The DTA published the full text on 7 August 2026. It is short, direct and worth reading in full if government work is part of your pipeline. The source is linked at the end of this article.
What the DTA actually said
A year ago, the public conversation was about when to speed up or slow down AI adoption. This speech marks the shift: the APS AI Plan has been released, and the task now is scaling AI across government without losing control of it.
The line that frames the whole speech:
"A pilot proves that something can work. Scale proves that we can operate it."
Lucy Poole, Digital Transformation Agency, 4 August 2026
Operating AI at scale, in the DTA's telling, means an agency must be able to sustain the cost, understand the architecture, manage the supplier, support the workforce and explain what happened when the system gets something wrong.
Two passages matter most for anyone on the selling side.
First, the DTA said government cannot scale AI alone. It needs industry to design for transparency, interoperability, cost visibility and changing assurance needs, rather than bolting them on after deployment.
Second, on what agencies will keep requiring as the pace of change increases:
"Agencies still need notification thresholds, audit trails, testing, rollback and the ability to pause when risk is unclear."
Lucy Poole, Digital Transformation Agency, 4 August 2026
And the relationship itself is changing. The DTA said partnership must extend beyond procurement into continuous operational assurance: visible changes, timely evidence and confidence that controls still hold. That is a description of an ongoing obligation, not a checkbox at contract signing.
The scaffolding behind the speech
The speech references a set of published instruments that now form the common baseline for AI in the Australian Public Service:
- The APS AI Plan, the shared direction for safe, responsible AI adoption across government
- The Responsible Use of AI Policy, which sets expectations for how agencies use AI
- The AI Impact Assessment Tool, used to assess AI use cases before and during deployment
- The AI Technical Standard, which sets technical expectations for AI systems in government
The DTA describes these as scaffolding that helps agencies ask the right questions earlier and explain decisions later. Vendors should read that as a preview of the questions that will reach them: through evaluation criteria, through contract clauses, and through the assurance conversations that follow award.
What this means if you bid for government work
A paragraph that says "we use AI responsibly" is not going to hold up much longer, because buyers are being handed tools that ask for specifics.
Based on the direction set in this speech, here is what an agency is likely to want to know about the AI in your offer, whether that AI is part of the service you deliver or part of how you prepared the bid:
| What buyers will ask | What you need to be able to show |
|---|---|
| What does the AI actually do? | A plain-English description of the AI's role in the workflow, and where its outputs go |
| What does it draw on? | The data and sources behind outputs, and how confidential or buyer-provided material is handled |
| Who checked it? | Named human review and approval before outputs are relied on or submitted |
| What changed, and when? | A record of versions, updates and who signed off, kept as the work happens |
| What does it cost to run? | Visibility of consumption and unit costs, not just a licence price |
| Can it be stopped? | Testing, rollback and the ability to pause the system when risk is unclear |
Compiled from the expectations stated in the DTA speech of 4 August 2026. Always check the specific requirements in each RFT.
None of this is exotic. It is the same discipline agencies are being asked to apply to themselves, extended to the suppliers they depend on. The vendors who feel this first will be those selling AI systems. But the bid process is not exempt, because a tender response is a supplier's first demonstration of how it works.
What this means for AI in your bid process
If your team uses AI to help prepare tender responses, the same questions arrive in a smaller frame. Which sources did the AI draw on? Did it use anything you cannot substantiate? Who reviewed each section before it went out? Is there a record?
Teams that can answer those questions have an advantage that compounds: evaluators reward bids they can trust, and a governed AI process produces the trust signals as a by-product of the work rather than as an afterthought.
The practical preparation looks like this:
- Ground AI output in your own evidence. Case studies, CVs, certifications and project records your business can actually produce, not general internet content.
- Keep human review in the loop, and record it. Every section reviewed and approved by a person before submission, with the approval captured.
- Keep a record of generation. What was generated, from which evidence, and who signed off, retained at proposal level.
- Be ready to disclose. A clear, factual statement of what AI was used for, what it was not used for, and how buyer information was handled.
For a deeper treatment of where AI disclosure in Australian bids is heading, see our guide to AI governance in proposal writing.
How Doreva approaches this
Doreva was built on a bet that this is where government buying was heading, before the frameworks named it.
Every material claim in a generated response traces back to the bidder's own approved evidence. A human reviews and approves every section before export, and nothing goes out on autopilot. The proposal-level record of what was generated, from what evidence, and who signed off is kept as the work happens, and Doreva's AI governance exports turn it into a document a buyer can read.
That matters because the alternative is reconstructing the story after the buyer asks, from memory and email threads. Bid teams should not have to choose between speed and defensibility.
FAQs
What did the DTA say about AI at scale in August 2026?
In a speech delivered on 4 August 2026 at the Tech in Gov event, Lucy Poole, Deputy CEO of the Digital Transformation Agency, said the conversation has moved from AI pilots to whole-of-government scale under the APS AI Plan, and that government needs industry to design for transparency, interoperability, cost visibility and changing assurance needs rather than bolting them on after deployment.
What is the APS AI Plan?
The APS AI Plan is the Australian Public Service's shared direction for safe, responsible AI adoption across government. It sits alongside the Responsible Use of AI Policy, the AI Impact Assessment Tool and the AI Technical Standard.
What will government buyers expect from vendors that use AI?
Expect questions about what the AI does, what data and sources it draws on, what a human reviewed and approved, how changes are notified, what the running costs look like, and whether the system can be tested, rolled back or paused. The DTA has said partnership must extend beyond procurement into continuous operational assurance.
Does this affect AI used to write tender responses?
Increasingly, yes. Buyers being told to govern their own AI use tend to ask suppliers the same questions. A bid team using AI should be able to say what it was used for, which evidence it drew on, who approved each section, and how buyer-provided information was handled.
Be ready before the buyer asks.
Start the free pilot with a live tender. Doreva reads the pack and shows the requirements and evaluation criteria first. When you draft, every claim traces to your own approved evidence, a human signs off on every section, and the governance record builds itself as you work.
Start Free PilotBuilt for Australian Government tenders. Human review before every export.
Sources: Digital Transformation Agency, "Speech: Moving from AI pilots to whole-of-government scale", delivered 4 August 2026, published 7 August 2026. Quotes are from the published text, which the DTA notes should be checked against delivery. This article reflects the speech as published, as at 10 August 2026. Always check the current requirements in the specific RFT before relying on any expectation described here.