
Insight18 November 2026
AI-Driven Attacks Are Rewriting Australia's Corporate Security Playbook
Australia's mid-market is confronting enterprise-scale cyber losses without enterprise-scale security capacity. A more effective response will go beyond compliance to make trustworthy automation an integral feature of the systems on which people and AI agents depend.
In brief / 10 min read
Australia's mid-market is confronting enterprise-scale cyber losses without enterprise-scale security capacity. A more effective response will go beyond compliance to make trustworthy automation an integral feature of the systems on which people and AI agents depend.
Key Takeaways
- The mid-market needs a different security model. Medium-sized Australian businesses face cyber losses close to those reported by large businesses, often with less specialist capacity. Safer systems, restricted access and tested recovery can do more to close that gap than asking small teams to absorb ever more work.
- Agentic AI expands both capability and exposure. Agents can help teams investigate, coordinate and act beyond the reach of fixed automation, but their access to data and tools can also be exploited. Organisations need governed environments, limited permissions and human oversight of consequential actions to realise the benefits safely.
- Assurance must be continuous and accountable. Leaders should use ASD Essentials to assess whether controls work across critical systems, while meeting existing obligations. Clear measures of identity, response, suppliers and AI use give boards evidence to guide decisions; insurance and outsourcing do not transfer responsibility for the underlying risk.
Why the Australian Mid-Market Needs a Different Security Model
The seventh Annual Cyber Threat Report 2025-26 records more than 82,000 cybercrime reports and 1,524 cyber security incidents handled by the Australian Signals Directorate (ASD), a 20% increase in incidents year on year. Average self-reported losses per business report rose to $93,500. For medium-sized businesses, the average increased 45% to $141,300, close to the $148,100 reported by large businesses.
These figures describe reported averages, not the probability or expected cost of an incident for any individual firm. Their strategic significance is nevertheless clear: the financial impact of a serious compromise does not scale neatly with the size of the organisation. A mid-market business can depend on the same cloud platforms, suppliers and digital channels as a larger enterprise while having far less capacity to investigate alerts or redesign access.
This challenges a familiar approach to security planning: additional products and annual training alone will not close a structural capacity gap. A more durable investment is to reduce the number of consequential decisions a small team must make under pressure through consolidated identity, narrow privileges, recoverable environments and evidence that controls cover the business.
AI makes this investment more urgent. ASD assesses that it lowers barriers to finding and exploiting software flaws while enabling more convincing impersonation and phishing. A survey of 400 Australian cyber security leaders reported that two-thirds of firms had experienced at least one AI-related threat in the preceding year. Such survey responses are useful signals of perceived exposure, but they are not an official measure of the share of Australian breaches caused by AI.
The Qantas third-party customer service platform breach illustrates the broader problem of trust across organisational boundaries. Even where synthetic media played no part in an initial breach, AI can make subsequent impersonation attempts more persuasive. Exposed customer data, supplier access and apparently legitimate communications can therefore combine to create risks that persist long after the original compromise.
From Compliance Snapshots to Operational Assurance
ASD launched ASD Essentials in October 2026 to broaden its baseline guidance across enterprise IT, cloud and operational technology. It organises guidance around principles, actions and domains, reflecting environments in which business functions depend on more than a fixed set of technical mitigations.
The framework's practical value will depend on how organisations apply it. Even an approved, purchased and documented control can leave an unmanaged identity, legacy application or supplier connection exposed. Leaders should therefore ask what proportion of the relevant environment is covered, whether the control works in practice, and how quickly exceptions are found and resolved. AI-enabled systems should be included in that scope rather than treated as an isolated experiment.
There is no mandatory ASD Essentials cutover deadline, and adoption does not automatically displace statutory or contractual obligations to report against the Essential Eight. Organisations should maintain those obligations while improving their ability to demonstrate effectiveness across the systems on which their critical work depends.
The insurance market provides another view of these same operational weaknesses. In the ESET survey, 95% of Australian leaders reported difficulty obtaining or maintaining cyber insurance, while 48% reported having coverage. Underwriting scrutiny can surface gaps in identity, recovery and supplier controls; it cannot make those controls effective. Evidence built for operational decisions can also support insurance discussions, rather than creating a separate assurance exercise.
Agentic AI: Productivity Gains and Faster-Moving Threats
Agentic AI gives teams new ways to coordinate work, investigate problems and carry out complex tasks. Unlike traditional automation, which generally follows a prescribed sequence, agents can interpret information, select tools and adapt their actions as circumstances change. Used well, they can reduce routine workload, improve productivity and extend specialist capability across an organisation.
The same characteristics also change the threat environment. Bad actors can use agents, including coordinated groups of agents, to search for weaknesses, adapt their approach and pursue multiple paths into a business at greater speed and scale. Misaligned agents can also take actions that run counter to an organisation's objectives, particularly when they have broad access to systems or operate without sufficient oversight. The result is not an entirely new set of security fundamentals, but a shorter window in which to find and address weaknesses before they can be exploited.
For Australian businesses, apparently minor gaps can therefore carry disproportionate consequences. An overlooked account, an overly permissive integration or an exposed component in the wider software architecture may become a route into more valuable systems. A security strategy that focuses only on individual AI tools misses the broader question of whether the surrounding environment is designed to withstand faster, more persistent probing.
The appropriate response is to improve the architecture that supports both productivity and resilience. Organisations should know which systems and identities they depend on, restrict access to what each user or agent needs, monitor important activity and make it possible to contain a compromise quickly. Approved AI tools, clear ownership and human oversight of consequential actions can help teams use agentic systems with confidence while reducing the opportunity for misuse.
Singapore offers a useful point of comparison. Its Cyber Security Agency has highlighted the use of agents in attacks and the risks to software pipelines, while its work on assessment and behavioural testing suggests a practical discipline: test how systems perform under realistic misuse and revisit those tests as models, permissions and integrations change. Australian organisations should adapt that discipline to their own systems, resources and obligations.
ASD's Opportunities for AI in Cyber Defence points to international resources including OWASP, MITRE ATLAS and the NIST AI Risk Management Framework. These can help teams examine plausible attack paths and make better design decisions. Their value depends on translating guidance into effective controls across the systems that support the business.
Accountability Remains With Australian Leaders
Australia's legal and regulatory settings make cyber resilience a governance issue as well as a technical one. Depending on the organisation and circumstances, obligations may arise under the Cyber Security Act 2024, the Security of Critical Infrastructure Act 2018 (SOCI Act), privacy law and sector-specific standards. Relevant critical infrastructure entities must consider incident reporting and risk management obligations; APRA-regulated entities must also address operational resilience and information security under CPS 230 and CPS 234.
Effective governance also extends beyond meeting reporting deadlines. Directors and executives need evidence that important risks are understood, that exceptions have owners, and that decisions are followed through. AI can help summarise material and identify questions, subject to appropriate confidentiality and verification, but it cannot exercise a director's judgement or assume personal accountability. Likewise, a managed security provider can supply capability without taking ownership of the client's risk.
Workforce constraints make clear ownership especially important. Jobs and Skills Australia identifies persistent shortages across several cyber and adjacent technical roles. Mid-market leaders should prioritise the capabilities that change outcomes including identity architecture, cloud and supplier oversight, incident response and recovery, while developing internal staff and using external partners where specialist depth is needed. The objective should be a security model suited to the organisation's scale, with accountable owners for essential decisions and controls.
Six Shifts for the Corporate Security Playbook
- Prove coverage and effectiveness continuously. Map critical services, identities and assets to relevant ASD Essentials outcomes. Test controls and resolve exceptions while retaining Essential Eight reporting where required.
- Make identity resistant to impersonation. Expand phishing-resistant MFA and passkeys, restrict privileged access, monitor exposed credentials and independently verify high-impact financial or access changes.
- Design for detection, containment and recovery. Reduce unnecessary internet exposure, monitor critical systems, rehearse isolation and recovery, and use AI-assisted triage with appropriate human validation.
- Build governed environments for people and agents. Provide approved AI tools and clear data-handling rules. Give agents separate identities and least-privilege access, log their actions and require approval for consequential changes.
- Treat suppliers as part of the operating environment. Identify critical dependencies, limit and review third-party access, test supplier controls in proportion to risk, and prepare for a supplier compromise as well as an internal one.
- Connect board oversight to operational evidence. Assign owners, fund remediation and review trends in a small set of meaningful measures. Pair specialist support with internal capability so that decisions remain informed and accountable.
Metrics Every Team Should Monitor
A useful dashboard shows the denominator, scope, trend, owner and material exceptions for each measure. Where practical, teams should also segment results by critical service and supplier, since an improving aggregate can conceal an exposed system. Targets should reflect the organisation's risks, and the quality of the underlying data should be tested regularly.
Metric | What to measure | Why it matters |
|---|---|---|
MFA coverage | Percentage of in-scope human accounts protected by MFA, with phishing-resistant coverage shown separately. | Reveals authentication gaps across workforce, remote access and critical applications. |
Privileged account coverage | Percentage of identified privileged accounts under approved, monitored access controls; track exceptions and review frequency. | Shows whether the accounts with greatest authority are governed, including service and agent identities where applicable. |
Mean time to detect | Average elapsed time from incident onset to detection for incidents where both times can be established; report sample size and median alongside it. | Helps assess detection capability, while recognising that unknown onset times can bias the measure. |
Mean time to contain | Average elapsed time from detection to containment, using a defined containment point; report sample size and median. | Shows how quickly teams limit harm after discovery and whether response improvements are working. |
Vendor criticality coverage | Percentage of active vendors assessed and assigned a criticality tier, with the share of critical vendors subject to current proportionate reviews. | Makes unassessed dependencies visible and directs scrutiny to services whose failure would matter most. |
AI inventory completeness | Percentage of discovered AI applications, models and agents recorded in the approved inventory with an owner, purpose, data use and access permissions. | Exposes unmanaged use and missing ownership; compare the inventory with procurement, cloud and identity records. |
Passkey adoption | Percentage of eligible users enrolled and actively using passkeys for supported services. | Tracks progress toward phishing-resistant sign-in; distinguish enrolment from actual use and note unsupported systems. |
These measures provide indicators rather than guarantees. A low reported detection time, for example, may reflect incidents that were easy to find rather than a comprehensive view of all intrusions. Teams should therefore review the measures together, investigate exceptions and periodically validate their conclusions through exercises.
References and Further Reading
- Australian Signals Directorate (ASD) / Australian Cyber Security Centre (ACSC). Annual Cyber Threat Report 2025-26 (October 2026). https://www.cyber.gov.au/publication/annual-cyber-threat-report-2025-26/year-in-review
- Australian Signals Directorate (ASD). ASD Essentials: Evolving Our Approach to Cyber Security Resilience (15 October 2026). https://www.asd.gov.au/news/2026-10-15-asd-essentials-evolving-our-approach-cyber-security-resilience
- Australian Signals Directorate (ASD). Opportunities for AI in Cyber Defence (September 2026). https://www.cyber.gov.au/publication/opportunities-for-ai-in-cyber-defence
- Office of the Australian Information Commissioner (OAIC). Large-Scale Data Breaches on the Rise, Latest OAIC Statistics Reveal (October 2026). https://www.oaic.gov.au/news/media-centre/large-scale-data-breaches-on-the-rise,-latest-oaic-statistics-reveal
- Hindsight Legal. OAIC Data Breach Statistics 2026: Large-Scale Breaches Rise (October 2026). https://www.hindsightlegal.com.au/oaic-data-breach-statistics-2026
- Thomsons Lawyers. Critical Infrastructure and Cybersecurity Risk: What Directors and Officers Need to Know (July 2026). https://www.thomsons.com.au/insights/critical-infrastructure-and-cybersecurity-risk-what-directors-and-officers-need-to-know
- Insurance News (reporting on ESET). Business Survey Reveals AI Breach Threat, and Gaps in Security (November 2026). https://www.insurancenews.com.au/insurtech/business-survey-reveals-ai-breach-threat-and-gaps-in-security
- Jobs and Skills Australia (JSA). Cyber Security Skills in Demand as Labour Market Evolves (2026). https://www.jobsandskills.gov.au/news/cyber-security-skills-demand-labour-market-evolves
- Cyber Security Agency of Singapore (CSA). Singapore Cyber Landscape 2025/2026 & CSA’s Initiatives to Strengthen Singapore’s Cyber Defences Amid an AI-Driven Threat Landscape (2026). https://www.csa.gov.sg/resources/publications/singapore-cyber-landscape-2025-2026/
- Infocomm Media Development Authority (IMDA). Artificial Intelligence in Singapore — Agentic Testing Starter Kit (2026). https://www.imda.gov.sg/about-imda/emerging-technologies-and-research/artificial-intelligence
- CrowdStrike. 2026 Global Threat Report (2026). https://www.crowdstrike.com/en-au/global-threat-report/
- IBM. Cost of a Data Breach Report 2026 — Shadow AI Has Reached the SOC (July 2026). https://www.ibm.com/think/insights/shadow-ai-has-reached-soc-security-teams-blind-spots
- Brandefense. What Is Agentic Security? A Plain Guide (OWASP Top 10 for Agentic Applications 2026 & MITRE ATLAS) (2026). https://brandefense.io/knowledge-hub/agentic-security/what-is-agentic-security/
Author
Teratek

A conversation is a good start
From insight to
impact.
Tell us what is getting in your team’s way. We’ll listen, ask questions and work out a practical next step together.
Continue reading
Explore more news and insight
Visit journal