Not sensitive? Less caution in new policies and guidance on public Gen AI use in Government

The Commonwealth’s Digital Transformation Agency (DTA) has issued new guidance on the use of public generative AI tools for Australian Government agencies and staff. This guidance replaces previous interim guidance, and represents a shift from a cautionary position to a risk-based approach that encourages controlled access to public generative AI and explicitly allows its use with OFFICIAL information (but not OFFICIAL: Sensitive or above).

The guidance should be read alongside the Protective Security Policy Framework Policy Advisory 001-2025 on OFFICIAL Information Use with Generative Artificial Intelligence, recently released by the Department of Home Affairs, which:

• provides agencies and stakeholders with confirmation that OFFICIAL information—including data used for business and service delivery—can be used with generative AI (subject to any policies covering the PSPF entity), and
• requires entities to only allow access to gen AI products hosted on providers certified under the Hosting Certification Framework OpenAI and Anthropic (or hosted by a provider that has undertaken a Foreign Ownership and Control assessment).  

This alert summarises these developments and outlines key takeaways, including steps to take now.

Time to reap the benefits

The DTA’s new guidance for agencies and staff[1] reflects the Australian Government’s controlled enablement approach to public generative AI, seeking to maximise its benefits by integrating it into agency work while managing its risks and limitations. The guidance documents operate at an enterprise level (ie need to be embedded into systems and processes) and a personal level (ie individuals need to comply), respectively.

Gen AI is here, and it is here to stay

Gen AI is now widely available and increasingly integrated in tools and services we use every day, often without explicit user notification. This includes via search engines, productivity applications and software platforms. Even if someone is not seeking to use Gen AI, a simple Google search will often prompt an ‘AI Overview’ to pop up. Because AI is often embedded in tools that are already used, its presence is not always obvious.

Australian Government agencies restricting access to Gen AI has not removed demand or interest in these tools. Rather, it means that staff:

1. may use these tools without proper training or security safeguards, and
2. lack practical experience using a technology that is becoming increasingly ubiquitous in professional and personal lives.

What does the DTA’s new guidance mean for public AI use in Government?

The DTA has responded to the proliferation and potential of public generative AI by providing a clearer, more enabling direction for its use in Government.

Public generative AI refers to web-based public generative AI tools. Public generative AI is in contrast to enterprise-grade generative AI, which is deployed in controlled environments to meet specific governance standards, and security and data protection requirements.

Public generative AI products hosted by providers certified under the Hosting Certification Framework or hosted on OpenAI (such as ChatGPT) or Anthropic (such as Claude) are approved under the PSPF Advisory note. Otherwise, the provider must have undergone a foreign ownership control, or influence (FOCI) risk assessment.[2]

The DTA’s updated guidance recognises the benefits of public generative AI – faster research, development of cleaner drafts, and more thorough analysis of documents and data. However, it also acknowledges the potential for security and privacy risks associated with misuse of these tools.

The updated guidance reflects this balancing act, and provides a clear way forward, distinguishing between staff and agency responsibilities.

For staff, the guidance is simplified into three overarching principles:

1. protect privacy and safeguard government information
2. use judgement and critically assess generative AI outputs
3. be able to explain, justify and take ownership of your advice and decisions

For agencies, it recommends measures to manage access and include safeguards, such as monitoring and human validation requirements, while prioritising enterprise-grade AI for sensitive or classified material. The update to the DTA’s guidance elevates AI literacy as a core workforce capability, calling for training, competency frameworks and practical experience.

What does the PSPF Advisory note say?

Three key things entities (being agencies and accredited persons) must do:

• only provide access to generative AI products hosted on Hosting Certification Framework providers, OpenAI or Anthropic or which have undergone a Foreign Ownership, Control, or Influence (FOCI) risk assessment,
• ensure staff training includes guidance on handling security classified information when using generative AI, and
• follow the existing PSPF technology authorisation process and consider relevant Australian Signals Directorate Guidance when approving access to generative AI tools for use with OFFICIAL information.

The DTA guidance encourages agencies and staff to use public generative AI safely and responsibly. The PSPF Advisory note complements this guidance by providing mandatory guardrails that agencies must apply when enabling that use, in order to protect sensitive and classified government information.

Key takeaways for agencies and staff

What do you need to do now?

• While the DTA recommends agencies enable public generative AI use, it is up to agencies to decide whether to enable access to public generative AI on their systems.
• Agencies should review their current public generative AI-related policies in light of the guidance materials from the DTA – this covers both enterprise policies and those applying at an individual level.
• APS employees, contractors and consultants should confirm their agency’s policy position and approved tools, complete any required training and ensure they are verifying the AI tool’s outputs before any external release of work.
• Contracts with contractors and consultants should be reviewed to ensure that the clauses reflect the agency’s circumstances.
• Check that you have only provided access to generative AI products hosted on Hosting Certification Framework providers, OpenAI or Anthropic. If not, do you have a FOCI clearance?