AI is already running inside the enterprise. Is Australia ready?

Artificial intelligence is no longer sitting at the edge of enterprise experimentation. AI has quietly become part of everyday enterprise operations. It is drafting emails, summarising meetings, supporting customer interactions and increasingly taking autonomous actions across business systems. The competitive advantage is undeniable, but so is the security challenge. AI is expanding the attack surface faster than most organisations are expanding their ability to see, govern and secure it. Security teams should accelerate AI adoption, not slow it down.

Proofpoint's 2026 AI and Human Risk Landscape report shows that AI adoption in Australia has already moved well beyond pilot stage. Four in five (80%) organisations have deployed AI assistants beyond pilot, while almost three quarters (72%) are advancing autonomous agents. Yet security readiness has not kept pace. More than half (53%) of Australian organisations describe their AI security posture as catching up, inconsistent or reactive. Close to two in five (39%) have already experienced a suspicious or confirmed AI-related incident.

This is the gap that should concern security leaders. AI is not waiting for governance frameworks to mature. Security leaders in Australia are under more pressure to address key areas of concern.

To put it simply, AI is pressure testing every system and process.

AI has expanded the attack surface

For many years, cybersecurity strategies were built around familiar control points: email, endpoints, cloud applications, identities and data repositories. Those still matter. But AI is now connecting these environments in new ways, allowing risk to move across workflows at machine speed. In reality, how many employees in confidential meetings are actually recording them on personal devices, then upload them to unsanctioned AI apps?

The challenge is no longer simply preventing compromise. It is maintaining context across rapidly connected workflows so security teams can distinguish legitimate AI activity from malicious or unintended behaviour.

In Australia, email remains the most common AI-related threat vector, affecting 53% of organisations. But exposure now extends much further: SaaS and cloud applications at 46%, AI assistants or agents at 39%, and social and messaging platforms at 37%. Among organisations that have already experienced an AI-related incident, exposure rises sharply across every channel - 67% in third-party SaaS and cloud applications, and 62% involving AI systems directly.

This matters because enterprise work no longer happens in a single channel. A sensitive document may move from email into a collaboration platform, be summarised by an AI assistant, stored in a cloud application, and referenced by an autonomous workflow. Each step creates another point where data, identity and intent need to be understood.

Many organisations already have some forms of AI security controls, for example, monitoring shadow AI applications. However, the critical visibility is whether those controls can see across the connected environment how AI is actually being used.

Data Security and AI Security Are the Same Problem

One of the most common structural errors in how organisations approach AI security is treating it as a separate workstream from data security. It is not. They are facets of the same problem, and solving one without addressing the other creates compounding exposure.

The earliest AI security challenge was clear: employees were using consumer AI tools to process sensitive business information. In 2025, 63% of employees who used AI applications uploaded confidential company data, such as source code and customer records, to personal chatbot accounts. According to IBM's Cost of a Data Breach Report, shadow AI breaches cost an average of AUD$968800 more than standard security incidents, driven by delayed detection and difficulty determining the scope of exposure.

The second wave is more complex. As organisations moved to enterprise AI platforms - Microsoft Copilot, Salesforce Einstein, and others - the question became not whether data was leaving the organisation, but whether AI tools were accessing only the data they were supposed to. That is a data security problem expressed through an AI lens.

The third wave is real-time and agentic. Autonomous agents do not just respond to prompts. Similar to humans, they connect to external tools and MCP servers, acquire new capabilities, and act on data across connected systems. Understanding what an AI agent is doing requires capturing not just the prompt and response, but every tool call and downstream action in between. When security teams do not have visibility into what AI is connecting to and acquiring, they cannot tell the board they have it under control. Organisations should also consider AI model governance, prompt integrity, and lifecycle visibility alongside traditional data protection. Without understanding how AI systems are configured, connected and authorised, data controls alone will not provide sufficient assurance.

Gartner projects that by the end of 2026, up to 40% of enterprise applications will integrate with AI agents, up from less than 5% in 2025. It also predicts that by 2028, 25% of all enterprise GenAI applications will experience at least five minor security incidents per year, up from 9% in 2025. The risk is scaling faster than governance.

Security and data governance teams need a shared view: what data exists, who and what has access to it, and how AI agents are actually using it. Having a clear view of all your data is not fictional, and it should be the foundation of building a robust AI security for any organisation.

Tool Sprawl is Holding Security Teams Back

Fragmented security stacks are compounding the challenge. Almost all (97%) organisations in Australia say managing multiple security tools is at least moderately challenging, and 45% describe it as very or extremely difficult. Respondents cite operational cost pressures (45%), integration challenges (43%), and overlapping or redundant tools (39%).

When controls sit in separate systems, security teams lose time moving between dashboards, reconciling alerts and trying to connect activity across email, cloud, collaboration and AI systems. That delay matters when incidents can spread across workflows quickly.

Investigation readiness reflects this structural weakness. Only about one in four (28%) Australian organisations say they are fully prepared to investigate an AI- or agent-related incident. More than one in three (36%) report difficulty correlating threats across channels. As AI-related activity spans email, collaboration platforms and cloud systems simultaneously, the ability to reconstruct what happened depends entirely on having visibility across those connected environments - which many Australian organisations do not yet have.

As AI scales, security architecture and adopting a platform approach becomes a strategic priority. 56% of Australian organisations are actively pursuing vendor and tool consolidation, and 54% believe a unified platform is more effective than point solutions. This reflects a broader shift in thinking. Over the next 12 months, more than two thirds (68%) plan to expand AI protections, 58% intend to extend collaboration channel coverage, and 56% expect to move toward a unified platform approach. Organisations are recognising that AI security cannot be solved with isolated controls. It requires an architecture that can protect people, data and AI systems across every channel where work is actually performed.

AI adoption in Australia is not slowing down. The boards and CEOs driving it are right that falling behind carries real competitive cost. Security leaders who can demonstrate visibility across people, data and AI systems will be in a stronger position to accelerate AI adoption with confidence rather than slow innovation through uncertainty. That is what setting the pace looks like.