Op-Ed: Australia’s AI gold rush is creating a cyber security skills crisis

New research from Barracuda, conducted on global organisations, including in Australia, found that 38 per cent of security teams lack the skills or expertise needed to secure and govern the AI technologies being used across the business. Among organisations with 500 to 1,000 employees, that rises to 49 per cent.

The findings reveal widespread skills gaps in AI security and governance, especially among mid-market organisations and businesses embedding AI into email systems, operational workflows, and enterprise systems. For Australian businesses, this should serve as a warning.

Pressing the accelerator without first checking the brakes

The AI conversation has become dominated by productivity. How much time can AI save? How many employees can it augment? How quickly can an AI agent complete a process that previously required a person? These are legitimate questions. But security leaders need to ask what happens when that AI gets compromised, manipulated, or simply gets something wrong.

The problem also becomes considerably more serious when AI moves beyond generating content and starts taking autonomous action. An employee asking an AI tool to summarise an email is very different to an AI agent with access to that mailbox, a CRM, internal documents, and business applications.

Suddenly, AI is not just another piece of software. It is a new identity with potentially powerful permissions. Australia’s cyber authorities are increasingly recognising this shift. The Australian Signals Directorate has warned that agentic AI introduces security, governance, and accountability risks that differ from traditional software and generative AI, driven by its autonomous behaviour, interconnected systems, and ability to operate with potentially excessive privileges.

VIEW ALL

The old security playbook is not enough

For years, cyber security has been built around protecting applications, networks, endpoints, and human identities. AI complicates all four. An AI agent can interact with systems on behalf of a user. It can access data. It can make decisions. It can call other tools. It can potentially trigger actions without a human reviewing every step. That means the traditional question of “who has access?” is becoming much harder.

Cyber security leaders now need to ask: What does the AI have access to? What can it do? Who authorised it? Can that permission change? Can we see what it has done? And what happens when it behaves in a way nobody anticipated?

ASD’s latest guidance on agentic AI warns that autonomy, tool access and connectivity can introduce new security risks, including privilege escalation, unintended behaviour and cascading failures across interconnected systems. Its September 2026 guidance on agentic AI further highlights how the infrastructure connecting AI models to data, tools, and enterprise systems shapes security, governance, and operational risk.

The technology is evolving quickly. Security teams must evolve faster.

Australia’s AI skills problem is bigger than AI literacy

Australia is already facing a broader skills challenge as AI changes the nature of work. Jobs and Skills Australia found in its 2025 Generative AI Capacity Study that AI is likely to reshape Australian work and skills, with adaptability and the development of digital and AI capabilities critical to navigating the transition.

But teaching employees how to use an AI chatbot safely is not enough. The next generation of cyber security professionals needs to understand AI models, agents, identity, data flows, APIs, privilege management, and AI-specific attack techniques.

Security teams also need to understand the business processes into which AI is being embedded. In Barracuda’s research, 42 per cent of organisations that use AI to integrate email with workflows reported concerns about AI governance expertise.

An AI ban is a Band-Aid solution – it will not hold

Banning outside use of AI won’t work because employees are already experimenting. People will continue to use AI tools because they are useful. They will upload documents, summarise sensitive information, generate code, and find ways to automate repetitive tasks.

If organisations respond by simply banning AI, they risk creating an even bigger problem: AI that is being used without visibility, governance, or security controls. And shadow AI is far more consequential than its predecessor, shadow IT, as it’s a fundamentally different risk model. An unmanaged SaaS application might expose a database; an unmanaged AI agent could potentially access the database, interpret its contents and act on what it finds – with AI, the list is endless.

Organisations need visibility over which AI systems are being used, what information they can access, what permissions they have and what actions they can take. They need tighter identity and access controls, monitoring, data protection and clear accountability. Most importantly, they need to invest in the skills to manage all of this.

AI governance cannot be something a security team is handed after the business has already deployed the technology. Security needs to be involved before an AI agent is given access to the systems that run the business.

Australia’s AI opportunity is enormous. But there is a growing risk that organisations are measuring the speed of AI adoption while ignoring the speed of AI risk. The biggest AI security problem facing Australian businesses is that AI is becoming powerful faster than the organisations deploying it can learn how to control it.