There is no shortage of hype around artificial intelligence. Every vendor has an AI story; every boardroom wants an AI strategy, and every conference keynote promises transformation. But when you strip away the marketing and focus on what AI actually means for cyber security right now, three realities emerge that every organisation needs to confront.
I shared these at a recent presentation in Townsville, and the response confirmed what I have been seeing across our client base: these are not future problems. They are today’s challenges, and they demand today’s attention.
Reality 1: Your AI Agent Has an Identity, and It Will Do Exactly What Its Permissions Allow
When organisations deploy AI agents, whether for automation, customer service, code generation, or internal operations, they tend to focus on what the agent can do. The more important question is what the agent is allowed to do.
Every AI agent operates under an identity. That identity carries permissions, access tokens, API keys, and trust relationships, just like any human user or service account. The difference is that an AI agent will exercise every permission it has been granted without hesitation, without judgement, and without the contextual caution a human might apply.
If your AI agent has write access to production systems, it will write to production systems. If it can access sensitive data stores, it will access them the moment its logic determines it should. There is no pause for reflection, no gut feeling that something seems off.
This means identity and access management for AI agents is not optional, it is foundational. Organisations need to apply the same rigour to AI identities that they apply to privileged human accounts: least privilege, just-in-time access, continuous monitoring, and regular attestation. In many cases, they need to apply even more rigour, because an AI agent can act at a speed and scale that no human can match.
The practical takeaway is straightforward. Before you deploy any AI agent, ask yourself: if this identity were compromised or if this agent behaved in an unintended way, what is the blast radius? Then scope the permissions accordingly.
Reality 2: Attackers Are Already Using AI to Move Faster Than Your Current Defences Were Built to Handle
This is not a prediction, it is an observation. Threat actors are leveraging AI to accelerate every phase of the attack lifecycle. Reconnaissance that once took weeks can now be completed in hours. Phishing campaigns are more convincing, more personalised, and produced at greater volume. Vulnerability exploitation is being automated in ways that compress the window between disclosure and active exploitation.
But here is the nuance that often gets lost in the panic: the vast majority of these AI-augmented attacks are not succeeding against well-architected modern systems. They are succeeding against legacy environments. Outdated software, unpatched systems, flat networks, and applications that were never designed with a threat model in mind. Attackers are using AI to find and exploit the weakest links faster, and those weakest links are almost always the systems that have not kept pace with modern security engineering.
Organisations that have invested in well-designed software with solid controls, secure-by-design principles, strong authentication, proper segmentation, and continuous patching, are already mitigating a significant proportion of what AI-powered attackers are throwing at them. The fundamentals have not changed; the speed at which you will be punished for ignoring them has.
This is not an argument for complacency. AI is genuinely lowering the barrier to entry for sophisticated attacks, and the window between vulnerability disclosure and exploitation is shrinking to hours rather than days. But it is an argument for prioritisation. If your environment still contains legacy systems with known vulnerabilities, unmanaged access, or architectures designed twenty years ago, that is where your risk concentrates, and that is exactly where AI-augmented attackers are looking first.
Reality 3: Governance Is What Lets You Move Fast Without Losing Control
There is a temptation to view governance as the enemy of innovation, the bureaucratic overhead that slows down AI adoption. In reality, the opposite is true. Good governance is what gives organisations the confidence to move quickly with AI, because it provides the guardrails that prevent experimentation from becoming exposure.
Governance in the AI context means having clear policies around acceptable use, data handling, model selection, and output validation. It means establishing accountability for AI-driven decisions. It means building audit trails that allow you to understand what an AI system did, why it did it, and what data it used to reach that conclusion.
Without governance, organisations face one of two outcomes. Either they move fast and accumulate unmanaged risk that eventually materialises as a breach, a compliance failure, or a reputational incident. Or they move slowly because nobody has the authority or the framework to approve AI initiatives, and they fall behind both their competitors and the threat actors who face no such constraints.
The organisations that are getting AI right are the ones that have invested in governance early, not as a blocker, but as an enabler. They can say yes to new use cases quickly because they have already established the boundaries within which those use cases can operate safely.
Where to From Here
- Treat every AI agent like a privileged identity – Audit the AI tools and agents already operating in your environment. Map their permissions, review their access scope, and apply least-privilege principles with the same discipline you would for a global administrator account.
- Stress-test your detection and response against AI-speed threats – Run a tabletop exercise where the adversary operates at machine speed, phishing at scale, automated lateral movement, exfiltration in minutes rather than days. If your current playbooks cannot keep pace, you know where to invest next.
- Stand up an AI governance framework before you need one – Define acceptable use, establish approval workflows for new AI tools, and build the audit capability to understand what your AI systems are doing with your data. The organisations that do this now will be the ones that can say yes to innovation without hesitation.
