Artificial intelligence has been dominating boardroom discussions for the past two years. Most of those conversations have focused on opportunity. How can AI improve productivity? Where can it automate repetitive work? How can it reduce operating costs?
This week, the conversation changed.
Cyber security agencies from the Five Eyes alliance, including Australia’s Australian Signals Directorate (ASD), issued a joint warning urging organisations to treat Agentic AI as an emerging cyber security risk that requires governance at the highest levels of the business, not just within IT teams.
That’s significant.
Government cyber agencies rarely publish joint guidance unless they believe organisations need to change how they’re thinking about a technology.
This isn’t a warning against adopting AI. It’s a reminder that AI systems capable of making decisions, interacting with other systems and executing tasks introduce a very different risk profile from traditional software.
For Australian organisations investing heavily in AI, that’s a conversation that belongs in the boardroom.
AI Assistants Are Becoming AI Operators
Until recently, most organisations used AI as an assistant. Employees asked questions. Generated documents. Summarised meetings. Created code. The human remained firmly in control.
Agentic AI changes that relationship.
Rather than simply generating content, these systems can:
- Complete multi-step tasks.
- Make decisions based on objectives.
- Access multiple business systems.
- Interact with APIs.
- Execute workflows automatically.
- Continue working without constant human input.
Think of it less as a chatbot and more as a digital employee.
That’s where the opportunity lies.
It’s also where the risk begins.
The more authority organisations give AI systems, the more important governance becomes.
If an AI agent can approve payments, access customer records, modify cloud infrastructure or trigger automated business processes, security teams need to start asking the same questions they would ask about any privileged employee.
Who approved this level of access? How is activity monitored? Can actions be reversed? What happens if the AI behaves unexpectedly? Those aren’t technology questions. They’re governance questions.
Why the Five Eyes Advisory Matters
The Five Eyes partnership has spent decades sharing intelligence on emerging cyber threats between Australia, New Zealand, Canada, the United Kingdom and the United States.
When those agencies publish joint guidance, it’s usually because they’re seeing the same risks emerge across multiple jurisdictions.
The recent advisory isn’t warning organisations to avoid AI. It’s encouraging them to implement AI securely from the outset. That’s an important distinction.
Every major technology shift has followed a similar pattern.
Cloud computing transformed infrastructure. Remote work transformed identity. Software-as-a-Service transformed business applications. AI is now transforming decision-making.
Each innovation delivered enormous business value. Each also created new attack surfaces that organisations needed to understand.
Agentic AI is no different.
The earlier organisations build governance around these technologies, the easier it becomes to manage risk as adoption accelerates. Waiting until dozens of autonomous AI systems are already operating across the business makes that task considerably more difficult.
The Risk Isn’t the Model. It’s What You Let It Do.
One misconception we hear regularly is that AI security is primarily about protecting the model itself.
That’s only one part of the picture.
From a security perspective, the more important question is: What can the AI actually access?
Consider two different scenarios. In the first, an AI assistant helps employees summarise meeting notes. It has no access to production systems, customer databases or privileged credentials. If it fails, the impact is relatively limited.
Now consider an AI agent connected to:
- Microsoft 365
- CRM platforms
- HR systems
- Financial applications
- Cloud infrastructure
- Internal knowledge bases
It can retrieve information. Generate reports. Update records. Create new users. Trigger workflows. Approve actions.
The conversation changes completely.
The AI itself may function exactly as designed. The risk comes from the permissions it’s been granted. That’s why many security leaders are beginning to view AI agents as another form of privileged identity. The same principles apply.
Least privilege. Strong authentication. Monitoring. Logging. Access reviews. Separation of duties.
Those concepts aren’t new. The technology they’re protecting is.
This Isn’t Just an IT Problem Anymore
Historically, most technology decisions sat comfortably within IT.
Today, AI projects are being driven by every part of the business. Marketing teams are experimenting with AI-generated content. Finance departments are evaluating AI for forecasting. HR teams are exploring recruitment assistants. Legal teams are testing document review. Operations teams are automating workflows.
That’s exciting.
It also means AI adoption is happening faster than governance frameworks can often keep pace. Security teams may not even know which AI platforms are being evaluated across the organisation. That’s why boards need visibility. Not because they need to understand every technical detail. Because they need confidence that appropriate governance exists before AI systems receive access to sensitive information or critical business processes.
The conversation is no longer:
“Should we use AI?”
It’s:
“How do we use AI safely, responsibly and with appropriate oversight?”
The Questions Every Board Should Be Asking
One of the reasons the Five Eyes advisory is so significant is that it shifts AI governance beyond the technology team.
Traditionally, boards have relied on executives to manage cyber security controls while focusing on strategic risk. AI is beginning to blur those lines.
An AI system that can access sensitive information, interact with customers or make operational decisions introduces risks that extend beyond cyber security. It can affect compliance, privacy, reputation, intellectual property and business continuity.
That doesn’t mean directors need to become AI experts. It does mean they should be asking the right questions.
For example:
- What AI platforms are currently being used across the organisation?
- Which systems do those AI tools have access to?
- Who approves new AI implementations?
- How are AI-generated actions monitored?
- What controls prevent sensitive information being shared with public AI models?
- Do we know where our data is going?
These aren’t technical questions. They’re governance questions. And they’re becoming just as important as asking about financial controls or business continuity.
Shadow AI Is Growing Faster Than Most Organisations Realise
Most organisations already have an AI strategy. The problem is that employees often have one too. It’s becoming increasingly common for staff to experiment with public AI platforms to improve productivity.
That might include:
- Uploading documents for summarisation.
- Generating software code.
- Analysing spreadsheets.
- Drafting contracts.
- Reviewing customer communications.
None of these activities are inherently unsafe.
The risk comes when employees unknowingly upload confidential information into systems that haven’t been approved or assessed by the organisation.
This is often referred to as Shadow AI.
It’s very similar to the rise of Shadow IT a decade ago.
Business users adopt technology because it helps them work more efficiently. Security teams only become aware of it much later. By then, sensitive information may already have been shared outside the organisation. That’s why visibility is becoming one of the most important aspects of AI governance. You can’t manage the risks associated with AI if you don’t know where it’s being used.
AI Doesn’t Remove Risk. It Changes It.
One mistake organisations sometimes make is assuming AI introduces entirely new security problems.
In reality, many of the risks are familiar. They’re simply appearing in different ways. For example, organisations already understand the importance of:
- Identity and access management.
- Privileged access.
- Data classification.
- Change management.
- Logging and monitoring.
- Third-party risk.
AI doesn’t replace these disciplines. It makes them more important.
An AI agent with broad permissions should be governed in much the same way as a privileged user. If an AI platform is integrated with business applications, it should be treated as another third-party supplier. If AI can access sensitive customer information, privacy obligations don’t disappear simply because a machine is processing the data.
Strong cyber security fundamentals remain the foundation. AI simply increases the importance of applying them consistently.
Red Teaming AI Is Becoming a Critical Security Practice
Traditional penetration testing focuses on identifying vulnerabilities in applications, infrastructure and networks. AI systems require a different approach.
Instead of asking whether an application can be compromised, organisations need to understand whether an AI model can be manipulated into behaving in unintended ways.
Examples include:
- Prompt injection attacks.
- Sensitive data extraction.
- Circumvention of built-in safety controls.
- Manipulation of automated decision-making.
- Abuse of connected tools and APIs.
These aren’t theoretical risks.
As AI adoption accelerates, security testing is evolving alongside it. AI red teaming allows organisations to understand how an attacker might attempt to influence or abuse an AI system before it’s deployed into production. This is particularly important where AI is integrated with customer-facing services, business workflows or critical operational systems.
Testing AI after an incident is considerably more expensive than validating it before deployment.
Security Should Be Built Into AI From Day One
One of the advantages organisations have today is that AI adoption is still relatively early. That creates an opportunity to build governance into projects from the beginning rather than retrofitting controls later. Security teams should be involved before AI systems are deployed, not after.
That includes reviewing:
- Data sources.
- User permissions.
- API integrations.
- Authentication methods.
- Logging capabilities.
- Monitoring requirements.
- Human approval points.
- Business continuity considerations.
The goal isn’t to slow innovation.
It’s to ensure AI delivers value without introducing unnecessary risk. The organisations that get this balance right will be able to adopt AI with greater confidence while avoiding many of the governance challenges that emerge when security is treated as an afterthought.
What We’re Seeing Across Australian Organisations
The level of interest in AI has increased dramatically over the past twelve months.
Almost every organisation we speak with is exploring how AI can improve productivity or automate business processes. What’s less common is a mature governance framework sitting behind those initiatives.
In many cases, AI adoption is happening faster than policy development. Individual business units are trialling new tools. Developers are integrating AI APIs into applications. Employees are experimenting with public AI platforms.
Meanwhile, security teams are still trying to understand what has already been deployed. That doesn’t mean organisations are doing the wrong thing. It simply reflects how quickly AI is evolving.
The challenge now is making sure governance catches up before AI becomes deeply embedded across the business. The organisations that move fastest over the next few years won’t necessarily be those adopting the most AI. They’ll be the ones adopting it with the strongest governance, visibility and security controls in place.
AI Governance Is Becoming a Competitive Advantage
One of the misconceptions around AI governance is that it’s purely about reducing risk. In reality, organisations that establish strong governance early often move faster than those that don’t.
Why?
Because they know where AI is being used. They understand what data it can access. They’ve defined who owns it. And they’ve built security into the deployment process from the beginning.
Without that foundation, every new AI initiative becomes a debate about risk. With it, organisations can adopt new technologies with greater confidence because the governance framework already exists. That’s why we’re seeing AI governance evolve from a compliance exercise into a business enabler.
The organisations that invest in it now are likely to be the ones best positioned to scale AI securely over the coming years.
AI Shouldn’t Bypass Your Existing Security Controls
One of the easiest mistakes organisations can make is treating AI as something separate from the rest of their technology environment.
It isn’t.
AI systems still connect to identities. They still consume data. They still interact with applications. They still rely on cloud infrastructure. That means they should be governed by the same cyber security principles already applied across the business.
Before deploying any AI solution, organisations should consider:
- Does the AI have more access than it actually needs?
- Is sensitive information being shared with external platforms?
- Are AI-generated actions logged and auditable?
- Can inappropriate behaviour be detected quickly?
- Is there a clear approval process before AI performs high-risk actions?
- Can access be revoked immediately if required?
If those questions haven’t been considered, AI may be introducing unnecessary risk without anyone realising it. Good AI governance isn’t about creating an entirely new security programme. It’s about extending existing security practices to a rapidly evolving technology.
A Practical Starting Point for Organisations
For many organisations, AI adoption has happened organically.
Different teams have trialled different platforms. Developers have integrated AI APIs. Business units have experimented with automation. Before investing in new tools, it’s worth understanding where you are today. A practical starting point might include:
Build an AI Inventory
Identify every AI platform currently being used across the organisation.
Include:
- Public AI tools.
- Commercial AI platforms.
- AI embedded within existing software.
- Internal AI projects.
- Third-party AI services.
Most organisations are surprised by how many AI-enabled applications are already in use.
Understand What Each AI System Can Access
Not every AI platform carries the same level of risk.
Focus on understanding:
- What data it can read.
- What systems it connects to.
- What actions it can perform.
- Whether it stores prompts or uploaded information.
- Who has administrative control.
This quickly highlights where additional governance may be required.
Review Permissions
One principle continues to apply regardless of the technology.
Least privilege.
If an AI system only needs read-only access, don’t grant write permissions.
If it only needs one application, don’t connect it to ten.
Reducing unnecessary permissions reduces potential impact if something goes wrong.
Test Before You Trust
Just because an AI platform functions as expected during demonstrations doesn’t mean it will behave the same way in production.
Security testing should become part of the deployment process.
That includes reviewing:
- Prompt injection risks.
- Data leakage.
- Access controls.
- API security.
- Authentication.
- Integration security.
Testing early is considerably easier than responding to an incident after deployment.
The Conversation Is Just Beginning
AI adoption is moving at a pace few organisations have experienced before.
New models appear almost weekly. Software vendors continue embedding AI into existing products. Business units are finding new ways to automate work.
That’s unlikely to slow down.
If anything, the pace will continue to accelerate. The challenge for security leaders isn’t stopping AI. It’s making sure governance evolves at the same speed. The organisations that succeed won’t necessarily be the ones using the most advanced AI. They’ll be the ones that understand where AI is being used, what risks it introduces and how those risks are being managed.
Conclusion
The Five Eyes advisory isn’t suggesting organisations slow down their AI adoption.
It’s encouraging them to approach it with the same discipline they’ve applied to cloud computing, identity security and digital transformation over the past decade. Agentic AI has enormous potential to improve productivity, automate business processes and support better decision-making. However, the more authority organisations give AI systems, the more important governance, visibility and security become.
Boards don’t need to understand every technical detail of AI.
They do need confidence that appropriate oversight exists, sensitive information is protected and AI systems are operating within clearly defined boundaries. As AI becomes embedded across more business functions, those conversations will increasingly become part of normal enterprise risk management rather than standalone technology discussions.
How Intrix Can Help
At Intrix, we help organisations adopt emerging technologies securely without slowing innovation.
Our consultants work with businesses to understand where AI is being used, assess the security implications of AI deployments and identify practical ways to reduce risk before issues become incidents.
Our AI and cloud security services include:
- AI Security Assessments to identify risks associated with AI platforms, integrations and workflows.
- AI Red Teaming to evaluate how AI systems respond to adversarial techniques such as prompt injection, data extraction and abuse of connected tools.
- Cloud Security Reviews to assess how AI services interact with cloud infrastructure, identities and sensitive business data.
- Security Risk Assessments to identify governance, compliance and operational risks associated with AI adoption.
- vCISO and Security Advisory Services to help organisations build practical AI governance frameworks aligned with business objectives and evolving regulatory expectations.
AI adoption doesn’t have to come at the expense of security.
With the right governance, testing and oversight, organisations can embrace AI confidently while protecting their people, systems and information.
If your organisation is exploring Agentic AI or expanding the use of AI across the business, Intrix can help you assess the risks, validate your security controls and build a governance framework that supports innovation without compromising resilience. Contact us today!
