Australia is building AI faster than it can secure it

Across Australia, AI is moving from pilot to production faster than most organisations expected. It is being embedded into customer service workflows, internal decision-making tools, marketing systems and software development pipelines. In many cases, it is already operating inside core business processes rather than sitting on the edge as an experiment.

More than half of Australian organisations are already using AI, yet only a fraction are operating with mature governance. Adoption has outpaced control, and most organisations are now building on systems they don’t fully oversee.

The speed is coming from inside the business. Product teams are integrating third-party models through APIs. Developers are using AI-assisted coding tools to accelerate delivery. Business units are deploying their own solutions to solve immediate problems, often without waiting for central approval. This is how adoption is actually happening on the ground.

Security and governance are not moving at the same pace: in most organisations, oversight, if there is any, still sits in central functions that were designed for a different operating model. Review cycles are slower. Visibility is partial. The assumption is that risk can be assessed before deployment, when in practice many AI-enabled systems are being deployed first and reviewed later, if at all.

This creates a gap that is easy to miss if you are looking at the wrong indicators. From a board or executive level, it can appear that AI adoption is being managed. Policies are in place. Frameworks are drafted. Risk committees are being updated. On paper, the structure exists but what is needed is actionable insight in real time.

Underneath all of this, the environment is shifting quickly. AI systems rely on access to data, integration with existing applications and the ability to trigger actions across systems. Each of those dependencies expands the attack surface. Data that was previously segmented is being exposed to new interfaces. Identities are being created for services, models and automation layers that don’t follow the same governance patterns as human users. Cloud environments are being reconfigured to support new workloads, often with broad permissions to keep things moving.

These are standard implementation choices made under delivery pressure.

The risk is not coming from some entirely new category of AI-specific vulnerability. It is coming from the way AI connects into systems that were already only loosely understood. Most large organisations already carry exposure across identity management, access controls and cloud configuration. AI adoption leans on those same systems and increases the speed at which weaknesses can be discovered and used.

There is also a structural issue that does not get enough attention: in many Australian organisations, accountability for AI risk is unclear. Technology teams are responsible for deployment. Security teams are responsible for risk. Data teams manage the inputs. Legal and compliance functions are often asked to interpret obligations after the fact. When something goes wrong, the question of ownership becomes difficult to answer.

This matters because the consequences are no longer theoretical. AI-driven processes are being used to make decisions that directly affect customers and the company’s reputation. Errors, misuse or compromise can move from a technical issue to a business issue quickly, whether that is through data exposure, service disruption or loss of trust.

Regulators are starting to respond, but regulation will lag deployment. This has been consistent across previous technology cycles. Waiting for clearer guidance is not a workable strategy when the systems are already in use.

What is required is a shift in how organisations think about exposure. AI cannot be treated as a separate category of risk with its own policy framework: it needs to be understood in the context of how access, identity and data already operate across the business. This requires visibility across environments that are currently managed in isolation, and a clearer link between technical exposure and business impact.

Some organisations are starting to adjust. They are focusing less on documenting AI use cases and more on understanding how those use cases interact with existing systems. They are looking at where access is expanding, where data is moving and how quickly changes are introduced. It is a more operational view of risk, grounded in how the business actually runs.

Many are not there yet. The pace of AI adoption is being set by competitive pressure and internal demand. Security is being asked to keep up with decisions it does not fully control.

This gap or lack of unification will define the next phase of cyber risk in Australia — not because AI changes everything, but because it exposes how much of the existing environment was never fully under control in the first place.

Image credit: iStock.com/Just_Super