Most organisations did not decide to start using AI. Their people did, one tool and one task at a time. The first job of AI assurance is not choosing a framework. It is finding out what is in use, what information it touches and who is accountable for it.
Use is running ahead of governance
The University of Melbourne’s 2025 global study, run with KPMG, includes a set of Australian findings. Around half of Australian employees (49%) say they intentionally use AI regularly at work, and 65% believe AI has been integrated into their organisation. Among employees who use AI, 48% admit to using it in ways that contravene company policy, including uploading sensitive company information into free public tools. Fifty-seven per cent have relied on AI output without checking its accuracy, and 59% have made mistakes at work because of AI. Yet only 30% say their organisation has a policy guiding generative AI use.
These are self-reported survey results, so they show the scale of a practice, not a count of incidents. They describe a pattern that most GRC teams will recognise: use arrives before policy, and policy arrives before verification.
What government and Defence guidance expects
Commonwealth. The Digital Transformation Agency’s Policy for the responsible use of AI in government applies to non-corporate Commonwealth entities. Version 2.0 took effect on 15 December 2025, with minor updates in version 2.1 on 21 October 2026. Its mandatory requirements include named accountable officials, transparency statements, use case accountability, internal use case registers, staff training and use case impact assessments. Businesses are not bound by the policy directly, but it shows what accountable AI use looks like to the government that buys from them.
The DTA’s staff guidance on public generative AI tools applies to employees, contractors and consultants working with government information. It tells people not to put information classified OFFICIAL: Sensitive or above, or personal information, into public tools, and to assume that anything entered could become public. It also points out that AI is increasingly built into everyday software, so it is not always obvious when it is active, and that look-alike tools can differ in how they handle data. Its own example is the consumer Microsoft Copilot versus Microsoft 365 Copilot.
For organisations outside government, the National AI Centre’s Guidance for AI adoption sets out six practices: decide who is accountable, understand impacts, manage AI-specific risks, share essential information, test and monitor, and maintain human control.
National direction. The Department of Industry, Science and Resources describes the National AI Plan (published 2 December 2025) as focused on capturing the opportunity, spreading the benefits and keeping Australians safe. The National AI Centre supports business adoption, and its Guidance for AI adoption builds on the earlier Voluntary AI Safety Standard and Australia’s AI Ethics Principles. Australia’s AI Safety Institute, which sits in the department and works with ASD and CSIRO, monitors and tests advanced AI to help government and regulators understand emerging capabilities and risks. Its focus is advanced and frontier AI. How your own organisation uses AI, and what safeguards it relies on, remains your responsibility.
Defence. Defence’s Policy Settings for Responsible Use of Artificial Intelligence in Defence commit Defence to lawfulness, values-based principles (including individual accountability, and AI that is reliable and secure) and proportionate, risk-based controls. They apply to the Department, the Australian Defence Force and the Australian Submarine Agency, not to industry. Defence’s public DISP cyber and assurance and how to apply pages do not set AI-specific requirements for members, so businesses in the supply chain should check their contracts and Defence Security Principles Framework obligations.
The Australian Signals Directorate’s AI considerations within the Essentials, published on 15 October 2026, gives the clearest technical direction. It says the Essentials apply whether an organisation consumes, develops or deploys AI. It treats AI systems as assets to identify and govern, including external AI services, connectors, and the accounts used by AI agents. It names sensitive data being provided to unapproved AI services among the risks.
Privacy. From 10 December 2026, the Office of the Australian Information Commissioner’s guidance on APP 1.7 to 1.9 expects privacy policies to describe certain automated decisions that significantly affect people and use their personal information. If your organisation uses personal information in AI-assisted decisions, take advice on what applies to you.
Why the information going in needs assurance
You are responsible for what AI produces, and people need to review it. The assurance question sits earlier: what goes in, and who and what can reach it.
The Digital Transformation Agency’s guidance notes that when people use public tools, their inputs and outputs may be shared with the tool provider. ASD’s AI considerations within the Essentials lists sensitive data being provided to unapproved AI services as a risk. It also describes indirect prompt injection, where instructions hidden in emails, documents or web content are processed by AI agents that have access to tools or data.
A policy statement does not address either risk. Both depend on controls that operate in practice: which tools are approved and enforced, what information they can reach, and what permissions agents hold. That is where independent assurance looks.
Five steps to start
- Find out what is in use. List the AI tools staff use, including features embedded in software you already own and personal accounts used for work.
- Decide what information can go where. Separate public tools from enterprise tools, and set clear rules for commercially sensitive information, client and contract details, tender material, source code and personal information. Treat anything entered into a public tool as potentially disclosed.
- Name accountability. Give each approved use case an owner, and name an executive who answers for AI use across the organisation.
- Check that the controls work in practice. Test who can access the tools and connected information, what information is exposed or retained, what permissions any AI agents hold, and whether approved-tool rules are enforced rather than only written down.
- Train, review and repeat. Train staff, require human review of AI output used for decisions, and re-check as tools, access and use cases change.
Five questions for executives
- Which AI tools are in use across the organisation, and who approved them?
- What information can they reach, and what rules govern commercially sensitive information?
- Who is accountable for each use case?
- How do we know our controls operate as intended, and who has checked them independently?
- What have we told our board, customers and regulators about how we use AI?
What AI assurance is
AI assurance is an independent assessment of whether your safeguards work in practice, supported by evidence. It does not review the quality of what AI produces, which remains your responsibility, and it does not certify a model’s accuracy. It does not replace your accountability or provide legal advice. You do not need to know which assessment you need before starting. A conversation about how you use AI, the information involved and the decisions ahead is enough to clarify a next step.
This article is general information, not legal advice. The official sources below set the requirements that apply to you.
Original ideas are Sypha’s. This article was drafted with AI assistance and verified against primary sources under our assurance process. Reviewed, edited and approved for publication by James Munro, Principal | Managing Director, Sypha Defence Advisory.
Related service: AI Assurance.
Sources
Checked 17 November 2026. These resources do not imply endorsement of Sypha.
- Digital Transformation Agency: Policy for the responsible use of AI in government
- Digital Transformation Agency: Using public generative AI tools safely and responsibly
- National AI Centre: Guidance for AI adoption
- Department of Industry, Science and Resources: Artificial intelligence
- Department of Industry, Science and Resources: Australia’s AI Safety Institute
- Defence: Policy Settings for Responsible Use of Artificial Intelligence in Defence
- ASD: AI considerations within the Essentials
- ASD: AI data security
- OAIC: New resources on transparency for use of AI and automated decision-making
- University of Melbourne: Trust and AI, the Australian story
- University of Melbourne: Trust, attitudes and use of artificial intelligence: a global study 2025 (full report)
- Defence: DISP cyber and assurance
- Defence: DISP how to apply
- Department of Industry, Science and Resources: National AI Plan