Mandatory AI safety and security protections for frontier AI: Australia sets its direction

The announcement sets out the Government’s intended approach at a high level and does not include draft legislation, so much will depend on how the obligations are drafted and on further consultation. Even so, the policy direction seems clear. The announcement is also notable for what it does not address, including limited use protections for information disclosed in incident reports, AI-related copyright issues and training data transparency.

What has been announced?

The Government states that the protections will be legislated through the AI Standards as part of its AI in Australia’s Interests agenda. Frontier AI is described as “the most sophisticated models at the cutting edge of current technological capability”.

A new targeted regulator, working with the Australian AI Safety Institute, would enforce mandatory obligations backed by “significant penalties” for non-compliance. The proposed regulator would have an enforcement role, with the Institute providing technical expertise and oversight as global standards evolve.

Under the proposed regime, frontier AI companies would be required to:

1. maintain security controls across all stages of development and deployment;
2. identify and manage serious risks associated with their models;
3. allow independent evaluation of their models;
4. provide mandatory notification of incidents to the Australian Government rapidly;
5. maintain and publish safety and security frameworks to ensure good governance, risk management and transparency; and
6. respond to updated standards and rules as technologies change.

As stated above, these requirements remain proposals at this stage. The announcement does not specify a commencement date or any specific thresholds for determining which companies or models are intended to fall in scope. We expect further details to be announced in due course.

The context: the Rapid Review into the OpenAI agent incident

On the same day, the Government released the recommendations of the Rapid Review into the OpenAI agent infiltration of an Australian Government health website, including the Medicare Statistics Reporting Service Portal.[1] The Government described the incident as unacceptable and the disclosure as inadequate. The cross-agency review, led by the Department of the Prime Minister and Cabinet, makes 18 recommendations, including consideration of mandatory, time-bound AI safety reporting covering misaligned behaviour and agentic cyber incidents. It also recommends exploring protections to support reporting of incidents, near misses and vulnerabilities, and consideration of criminal law reform in cases of serious misconduct. The Government has accepted the recommendations and says implementation work has commenced.

Read together, the two announcements suggest that incident notification will be one of the key components under the proposed regulatory regime, and that reportable events are likely to extend beyond conventional data breach notifications.

The Joint Select Committee’s early advice

The Government says the protections are “aligned with interim report recommendations from the Joint Select Committee on Artificial Intelligence”. The Committee provided early advice to the Prime Minister by letter dated 12 October 2026.[2]

Each of the six announced obligations reflects a recommendation in that letter. However, the Committee’s work is not complete, with the Committee still considering further evidence. The full report is due by 30 November 2026.

Some of its early recommendations are not referred to in the announcement, although given its high-level nature, that does not mean they will not be adopted.

Limited use protections

One recommendation from the Committee not expressly addressed in the safety and security announcement is to expand the “limited use protections”. These protections are intended to restrict the circumstances in which certain information about an unauthorised activity voluntarily disclosed by a frontier AI company can subsequently be used against it in enforcement action, thereby encouraging early and open reporting.

However, the Prime Minister’s latest announcement makes no mention of providing any limited use protections. That said, such protections may still form part of the eventual legislation. Recommendation 4 of the Rapid Review separately calls for exploring protections analogous to existing limited use protections for incident and vulnerability disclosures, a recommendation which the Government has accepted. However, how the protections would be legislated, if at all, will be a point to watch.

Copyright and training data: still to come?

The announcement also does not address AI-related copyright issues, including the use of copyright material to train AI models. That is perhaps unsurprising given the announcement’s safety and security focus.

The Joint Select Committee has noted that current arrangements are not facilitating licensing deals for Australian content and are seen by frontier AI companies as a barrier to training in Australia, recommending that the Government examine them.

The announcement is also silent on potential requirements to disclose a model’s training data, an approach previously floated by the Department of Industry, Science and Resources.

These matters may conceivably still be addressed in a later announcement or as one of the “other policy areas” flagged by the announcement as areas which the AI Standards legislation will cover.

The detail will be in the drafting

At this stage, the proposed requirements are statements of policy, and the Government has said further details will become available “in due course” after consultation. Many of the details remain to be settled, with key open questions including:

• how “frontier AI company” will be defined, and whether the test will turn on compute thresholds (such as a measure based on floating-point operations (FLOPs) adopted in other jurisdictions), model capability, revenue or some other measure;
• what the independent evaluation requirement will entail, including how the legislation will balance the need for meaningful independent scrutiny against frontier AI companies’ legitimate concerns regarding access to proprietary models and commercially sensitive information;
• what counts as a “serious risk” and as a notifiable “incident”, and how “rapidly” a notification must be made;
• the level of the “significant penalties”, whether they will attach to directors or senior officers, and whether limited use protections will apply to information provided in incident reports; and
• whether, and through what mechanism, AI-related copyright issues will be addressed.

What does this mean for organisations that use AI?

Most Australian organisations will not be frontier AI companies, and the obligations announced are not directed at them. However, the announcement demonstrates the increasing maturity of Australia’s approach to AI regulation and the growing regulatory scrutiny of how AI is developed, procured and deployed.

It is therefore a timely reminder for organisations to revisit their AI compliance and governance arrangements, including:

• mapping their use of AI and having clear visibility of the adoption of AI within their organisation;
• assessing whether their use of AI gives rise to any automated decision-making transparency requirements commencing on 10 December 2026;
• reviewing procurement processes and supplier contracts to account for AI-specific risks;
• reviewing incident response arrangements to ensure they address AI-related events (including AI-enabled cyber attacks);
• reviewing AI governance frameworks and ensuring appropriate board and senior management AI oversight; and
• considering training data, intellectual property and licensing issues arising from the adoption and/or development of AI tools.

Piper Alderman’s AI team combines expertise in technology, intellectual property, privacy, cyber security, disputes, governance and public policy to help organisations to adopt AI and stay ahead of fast-moving regulation.

[1]  Prime Minister the Hon Anthony Albanese MP, ‘Strengthening Australia against AI-driven cyber incidents’ (Joint Media Release, 10 November 2026) <https://www.pm.gov.au/media/strengthening-australia-against-ai-driven-cyber-incidents>.

[2]Joint Select Committee on Artificial Intelligence, Parliament of Australia, Letter to the Prime Minister (12 October 2026) (‘JSCAI Early Advice’) <https://static.aph.gov.au/-/media/Committees/ai_ctte/JSC_AI_letter_20261012_Final.pdf?rev=4d37b88a009c472d931e371b2d70d835&hash=1E01C0AF71D7362E4E1605A39BF2E654>.