As the integrity and procedural costs of AI systems rise with great velocity, Australia must take integral steps towards protecting its biological frontier.
Brussels writes the rules through its AI Act, the world’s first comprehensive horizontal statute classifying AI systems by risk. Washington picks the speed lane through executive orders, voluntary frontier model commitments, and federal procurement leverage rather than a single binding law. Beijing welds AI to industrial strategy, treating model development as state-directed infrastructure tied to national champions. Australia, on the Indo-Pacific edge of every AI supply chain, has neither the capital nor the compute to mimic any of them. It faces a sharper question: where to place the weight of its institutions when a technology moves faster than democracy can legislate.
Canberra has begun to answer. The National AI Plan, released on 2 December 2025, confirmed that Australia would not pursue a standalone AI statute, choosing instead to govern AI through the legal frameworks already in place. The Australian AI Safety Institute (AISI), funded at $29.9 million within the Department of Industry, Science and Resources, has joined the International Network of AI Safety Institutes alongside the United Kingdom, the United States, Canada, the European Commission, France, Japan, the Republic of Korea, Singapore, and Kenya. The question is whether these moves will hold weight when AI shifts from automating welfare decisions to engineering biology.
The challenge is not how to regulate AI, but how to govern a technology whose speed and opacity are testing the limits of democratic oversight: parliamentary scrutiny, judicial review, and the right of affected citizens to understand and contest decisions made about them.
The Speed Paradox
Australian lawmaking is slow by design, depending on consultation, evidence, drafting, and scrutiny. These are protections of procedural legitimacy, but they create exposure to technological speed. While Australian consultations run for months and bills move through parliament across sessions, frontier models ship major capability upgrades every few weeks. While regulators draft non-binding guidance, companies push systems into production for millions of users. While parliaments still debate definitions, AI is already embedded in workplaces, markets, public services, laboratories, and critical infrastructure.
This is the speed paradox. An AI Act in the Brussels style, a single horizontal statute that classifies systems by risk tier and imposes prescriptive obligations on providers and deployers, would deliver clarity for industry and citizens but risk rigidity as the technology outpaces the statute; voluntary principles would preserve flexibility for innovation but hollow out accountability when harms emerge. The National AI Plan chooses regulatory judo: govern the consequences of AI through systems already in place, including privacy, consumer, competition, and workplace law, and sectoral regulators from the OAIC and ACCC to eSafety, APRA, and ASIC. The approach avoids a single horizontal statute of the kind adopted in the EU. It also carries a structural weakness: a fragmented model that creates gaps, overlaps, and ambiguity, with no single regulator carrying the technical capability, mandate, or budget to govern frontier risk alone.
The Cost of Drift
That weakness already has a price. Robodebt remains the clearest national warning. The Royal Commission’s final report, delivered on 7 July 2023 by Commissioner Catherine Holmes, dissected a scheme that became a major failure of legality, fairness, and public trust. Its deeper lesson is that institutions become dangerous when efficiency is permitted to outrank scrutiny and procedural legitimacy.
That lesson now scales outwards. Automated systems already shape who receives welfare, who is shortlisted for employment, who is approved for credit, and who is flagged in migration. The OAIC’s determination against Clearview AI established that the body itself can become a profile, a credential, and a target.
From 10 December 2026, Australian Privacy Principle entities using personal information in automated decision making must disclose the kinds of information used and the kinds of decisions made. The reform does not ban black boxes, the opaque AI systems whose internal logic cannot be readily inspected or explained to the people they affect. It makes them harder to defend.
The deeper weakness is execution. Boards approve AI strategies they cannot interrogate. Executives endorse models without grasping the basic technical risks. Behaviour silently changes between versions. Outputs are fluent and confident but factually wrong. Hidden instructions buried in user inputs can hijack a model’s responses. And once a single provider is embedded across an organisation, switching away becomes prohibitively costly. The AISI is a meaningful step, but its mandate is advisory. Its main tools are pre-deployment evaluation, red teaming, and capability testing. The first tests models before they reach the public. The second stress-tests them for misuse by simulating adversarial users. The third measures what a system can actually do, including capabilities its developers may not have intended. None of these tools constrains risk unless the body running them has the authority to delay or refuse deployment.
The Biological Frontier
These are early problems. Harder ones lie at the biological frontier.
Biological design tools powered by AI can accelerate drug discovery, protein modelling, and disease research. DeepMind’s AlphaFold has predicted the structure of nearly every known protein, and a newer generation of generative models can design novel proteins and nucleic acid sequences from scratch. The OECD has flagged the convergence of synthetic biology, AI, and automation as a serious challenge for biosecurity, biosafety, supply chains, and human oversight.
Here Australia’s technologically neutral approach strains. Privacy law, consumer law, and workplace law were built for digital harms to people, not biological matter generated from models. None of them can screen a printed nucleic acid sequence or reach an open-source model running offline. When AI moves from information to matter, post-hoc disclosure stops being a remedy.
Where to Drop the Anchor
If Canberra cannot police every model, it must identify the points where digital velocity becomes physical capability. These are the new governance chokepoints: DNA synthesis providers, benchtop synthesis devices, biofoundries, cloud and supercompute infrastructure, public procurement, and critical infrastructure operators. The US Framework for Nucleic Acid Synthesis Screening offers one template. It governs the moment a digital sequence becomes physical DNA, not the moment it is designed on a screen. This matters because design tools proliferate beyond any regulator’s reach, but the equipment that turns sequences into matter, commercial providers and benchtop synthesisers, is a tractable chokepoint where screening can be enforced.
The organising principle is simple: the more consequential the system, the stronger the accountability around it. High-risk AI shaping rights, services, employment, credit, health, education, migration, or welfare needs transparency, contestability, and continuous human oversight. Frontier risk AI is different again. Where the technology touches biology, advanced compute, critical infrastructure, identity, or national security, safeguards have to be in place before harm, not after.
Australia will not match American capital, European regulatory scale, or Chinese industrial coordination. But it can decide where to apply institutional gravity: where AI affects rights, where automation shapes opportunity, and where digital designs become physical risks. As one of four Indo-Pacific members of the AI safety institute network alongside Japan, the Republic of Korea, and Singapore, Australia is well placed to export governance through the network rather than import risk through the supply chain.
Australia cannot stop the velocity. It can decide where to drop the anchor.
Muhammad Amir is a PhD researcher at Deakin University, specializing in international relations and security studies. His research focuses on peace processes, strategic competition, defence policy, and emerging technologies.
This article is published under a Creative Commons License and may be republished with attribution.
