Built-in Protection: How Australia's Public Safety History Can Guide the AI Era

The security challenges posed by AI can seem daunting - until you consider that Australia has faced and overcome many other significant challenges in the past.

Take the automobile. Australians fell in love with their cars through the 1950s and 1960s, but by 1970 faster vehicle speeds had pushed the annual road toll in Victoria alone to 1061. That epidemic demanded a response that went beyond what public education and traffic rules could provide.

Legislators recognised that road safety required combining human accountability with compulsory, engineered physical controls, and in 1970 Victoria made global history by becoming the first jurisdiction in the world to introduce mandatory wearing of seat belts. 

Today seat belts have been supplemented with airbags, crumple zones, and a host of safety features built directly into every car. It was this combination of engineered controls and driver training that saw Victoria's annual road toll decline by more than 70 per cent over the next three decades.

A familiar road

Today, managing the security risks of AI requires a similar shift, from relying on rules governing how technology is used to building protections into the technology itself.

Regulatory guidelines provide necessary boundaries, but they cannot stop an attack that tricks an AI system into following malicious instructions or halt a breach in progress.

This challenge becomes greater still when AI systems are given the ability to act autonomously across multiple government systems. From breaching public-facing sites to finding vulnerabilities in critical administrative platforms, Australia is witnessing the rapid evolution of software that acts autonomously and with authority. 

The fundamental governance question facing Australian leaders is no longer about controlling what software can see - it is governing what software is authorised to do.

The Australian Federal Government parliamentary inquiry into Artificial Intelligence has laid bare a stark reality - that acknowledging the risks of autonomous, rogue AI agents after a breach occurs is not sufficient. If Australia is to capture the full economic and social benefits of AI, we must do so without opening a dangerous operational security gap.

That requires moving beyond high-level governance promises and towards embedding "Secure AI by Design" principles into AI systems from the outset, supported by continuous, real-time technical safeguards.

The Unseen Threat: AI Supply Chains and Agentic Risks

Artificial intelligence might be the most significant new security challenge facing Australian organisations, but it also offers the most powerful new tools for their defence.

However, protecting AI systems - and protecting all systems from AI - requires organisations to first fundamentally rethink their security practices to counter these new threats. Autonomous AI agents can be tricked into revealing sensitive information, or to impersonate legitimate users, or even to take actions they were never authorised to perform. Traditional cybersecurity tools were not designed to manage these threats.

Crucially, as adversaries use AI to compress cyber-attack timelines from days down to minutes, manual approvals and periodic audits cannot keep pace.

The risks associated with AI agents are compounded by how they are built, as modern AI applications are rarely created entirely from scratch. Instead, they are assembled using models, data, and software components developed by numerous third parties, each of which can introduce its own security risks.

These factors mean AI models and the tools they use must be checked before deployment, with their behaviour continuously monitored as systems evolve.

Equipping Defenders and the Australian AI Safety Institute

Thankfully, the same speed and reasoning capabilities that make AI a potent tool for threat actors can also be turned inward to protect our digital infrastructure.

Under expert human supervision, advanced AI systems can help cybersecurity teams discover weaknesses, understand how attackers might exploit them, and rapidly put protections in place. Centralised security systems can also monitor and control how AI applications interact with organisational data and other systems, helping ensure their actions remain authorised and accountable.

This rapid defensive capability is a vital lifeline for critical infrastructure such as energy grids, transport networks, and healthcare systems, where operational systems cannot easily be taken offline for conventional patching.

However, defenders cannot mitigate risks they cannot analyse in advance. For Australia to stay ahead of machine-speed threats, government policy must facilitate early, managed access to cutting-edge frontier models for national defenders, security researchers, and critical infrastructure operators before those models are broadly deployed.

Australia's AI Safety Institute has an important role to play in evaluating emerging risks. But its role should also extend to helping ensure that government agencies, security researchers, and critical infrastructure operators can safely access advanced AI capabilities for defensive purposes.

Securing Australia's Digital Future

Just as the mandated use of seatbelts saw Australia lead the world in vehicular safety half a century ago, we now have the opportunity to set a global standard for secure AI adoption.

Australia can achieve the same balance between innovation and safety by ensuring security is built into AI systems from the outset. That means adopting continuous technical safeguards, giving defenders timely access to advanced AI capabilities, and ensuring those responsible for protecting critical systems can respond as quickly as attackers.

Done well, this approach will allow Australia to realise the benefits of AI while protecting national security and maintaining public trust. The recommendations from Parliament's inquiry provide an opportunity to ensure our governance frameworks and technical defences keep pace with the technology itself.