Government Agencies’ transparency statements are in – now it’s time to review them! Agencies’ attitudes to the statements vary significantly across the board, and varying degrees of transparency are achieved. A review of various statements shows some interesting trends and in our opinion the ATO, DCCEEW and Services Australia are some of the standout Agencies leading the way.
Welcome back to the fourth article in our AI in Government series.
An overview of Agency transparency statements
Only 60% of Agencies required to publish a transparency statement under the policy for the responsible use of AI in government (the Policy) did so before the 28 February 2025 deadline. As a reminder, all Non-corporate Commonwealth entities (other than National Intelligence Community Agencies) were required to publish an AI transparency statement.
Among the Agencies that issued statements, the level of detail provided varies significantly. Some Agencies categorised their AI use into the relevant usage patterns and domains, and list specific use cases. Others stated that they don’t use AI at all, which is interesting given the Policy’s broad definition of AI extends to non-generative AI tools that many agencies would already be using, including text-to-speech functions.
So, how does your transparency statement compare to that of other leading Agencies? How can you improve your transparency statement? And how can you monitor AI use in your Agency to ensure your statement is being updated appropriately?
Compliance with the DTA’s guidelines
The DTA’s Standard for AI Transparency Statements lists a number of key requirements that should be included in an Agency’s transparency statement. Now that the deadline has passed, we can see whether Agencies have met these key requirements and whether overall transparency has been achieved.
|
DTA Guidance
|
What we have observed
|
Example
uses 2
|
|
|
Types of AI addressed
|
Any AI system that falls within the OECD definition of AI should be included in the transparency statement. |
The OECD definition of AI includes more than just generative AI. Some Agencies recognise this explicitly, others demonstrate this understanding through the inclusion of non-generative AI use cases in their statements. Other Agencies seemed to adopt a narrower definition of AI, limited to generative AI. For example, one Agency uses text-to-speech functions on their website (a form of non-generative AI), but their transparency statement states that they don’t use AI. |
|
|
Use case examples
|
Agencies are to provide a “high-level overview of Agency AI use and management”. They are not required to list individual use cases. However, use cases and real examples can help to demonstrate how AI is being used and governed in the Agency, increasing transparency. |
Many Agencies list out individual use cases (eg the ATO has developed AI models to help reduce taxpayer compliance costs by using AI to prompt individuals to double check the amount they have entered for certain tax return labels when the amounts don’t match what is expected compared to other similar taxpayers). Others provide no use cases, including the DTA themselves who state that they don’t use AI in any way that interacts with the public. |
|
|
Identifying usage patterns and domains
|
“Agencies must list both the usage patterns and domains which apply to their use of AI in their transparency statements”. The usage patterns are using AI for decision making and administrative action, analytics for insights, workplace productivity, and image processing. The six domains are using AI for service delivery, compliance and fraud detection, law enforcement, intelligence and security, policy and legal, scientific, and corporate and enabling. |
Where Agencies identify use cases and recognised that they use AI in their statements, they tend to address usage patterns and domains as well. Where Agencies claim that they don’t use AI, they have not addressed usage patterns or domains. |
|
|
Training of staff
|
To report on compliance with the Policy, Agencies need to describe how they are training their staff. |
Majority of Agencies address their current staff training requirements. Many Agencies require staff to undertake some kind of training before they can access internal AI. Others failed to address staff training at all. Reminder: the Policy strongly recommends that Agencies implement the AI fundamentals training for all staff, regardless of their role, and that additional training is given to staff in consideration of their roles and responsibilities. The DTA confirms in its transparency statement that it has taken this approach to staff training. |
|
|
Explanation of governance structures
|
Transparency statements must include information about how Agencies are monitoring and governing the effectiveness of AI systems. |
Majority of Agencies at least state that they have internal governance arrangements in place for AI use. Some go into detail and describe their governance arrangements. Other Agencies, including Agencies that are already using AI, are still in the process of developing governance arrangements. |
|
|
Accountable officials
|
There is no explicit requirement for Agencies to publish who their accountable official is. They are however, supposed to be the primary contact for AI coordination, so it is useful if they are identified. |
Almost all Agencies list who their accountable officials are. Either by naming them or indicating a role in the Agency that will also be the accountable official (eg the COO will be the accountable official). Some agencies list more than one accountable official. This includes the DTA themselves who list two and explicitly divide the responsibilities between them. |
|
Improving transparency statements
Agencies should remain alert to the fact that they need to continue updating their transparency statements. They should do so every 12 months or when their use of AI significantly changes. When updating your statement, ensure you email a link of the updated statement to the DTA at [email protected].
We have come up with a number of key recommendations to help Agencies improve their statements and increase government transparency around AI use:
- AI use registers: Establishing an AI use case register can greatly assist Agencies when it comes to updating their transparency statements. It will contain specific use cases to draw on and allow the Agency to easily assess which usage patterns and domains they use AI for.
- Understand the scope of transparency statements: Agencies need to ensure they understand what falls within the OECD definition of AI. A lack of recognition of non-generative AI systems some Agencies’ transparency statement suggests they adopted a narrower definition to what constitutes AI.
- Look to other transparency statements: Comparing to top quality examples, like the ATO, DCCEEW or Services Australia, can give you ideas on how to improve your own transparency statement. Mirroring the way they list specific use cases, identify usage patterns and domains, address internal AI governance structures and report on staff training could help to improve your Agency’s statement, increasing transparency and consistency.
What should I do now?
Departments that have not released a transparency statement should do so now. Whilst the February deadline has been missed, it is important that Agencies still publish their statements in line with the DTA’s Guidelines.
If you have any questions or need help drafting or updating your transparency statement, please contact the KWM team.
