The Australian Signal Directorate (ASD) responded to more than 1,500 incidents last year — an increase of more than 20 percent over the previous year — as attackers used automation to exploit known weaknesses faster than poorly managed organisations could fix them.
What the Australian Signal Directorate reported in its Annual Cyber Threat Report 2025–26
In the ASD’s Annual Cyber Threat Report 2025–26, published in a 26 October report, the agency urges organisations to embrace AI for cyber defence but warns that asking “how much AI your cyber team is using” is the wrong question. The report stresses that no AI product replaces foundational knowledge: knowing what software you run, which of it is out of date, and what data you have needlessly exposed to the internet.
The ASD links the rise in incidents — and the speed with which attackers strike — to attackers’ increased use of automation. That, the agency says, makes basic weaknesses more dangerous because poorly managed organisations cannot patch or remediate quickly enough.
Why “AI cyber defence” is not one thing
The report emphasises that “AI cyber defence” describes a range of approaches rather than a single capability. Some uses of AI can be useful in narrowly defined circumstances: AI that automatically checks a system’s security settings helps organisations with weak IT configuration management, while AI that checks source code for vulnerabilities can add value for software developers with robust quality assurance processes.
Conversely, many organisations will gain little from source-code checking tools or from agentic AI that detects and responds autonomously unless they already have very mature processes, well‑understood networks and strong governance of AI itself. The ASD’s guidance is pragmatic: specific tools must be matched to specific weaknesses, and sometimes the best answer will not involve AI at all.
Your scanner finds 4,000 vulns. Which 12 matter?
Nubivance is a Rapid7 Registered Partner delivering vulnerability management as a service - scanning, risk-based prioritization, and remediation follow-through across IT and OT.
Fix the backlogCommon failure modes and practical risks of defensive AI
The report lays out failure modes that boards and security teams must consider before depending on AI tools. Large language models can hallucinate, producing confident findings about vulnerabilities that do not exist, and open-source software maintainers have publicly complained about floods of bogus AI‑generated bug reports. Organisations without the expertise to validate outputs risk chasing phantoms or, worse, trusting tools that miss real problems.
Other risks named by the ASD include sending security data to an AI service without controls on where that data goes and who can access it, and granting agentic systems broad permissions to act on networks — which can produce mistakes “at machine speed.”
Automation, patching and the basics still matter
ASD notes that automation long predates the current AI boom. Automated patch deployment, attack surface scanning and configuration baselines enforced by policy are “established, predictable technologies” that directly address weaknesses attackers exploit without necessarily needing AI. The report cites an example of attackers exploiting a vulnerability first identified in 2008, where regular software updates alone would have prevented compromise.
Similarly, the agency points to a decades-old Medicare statistics portal recently accessed by an OpenAI agent as an example of a system that did not require “smarter” defences so much as a simpler configuration change — taking it off the internet.
What this means for technologists, procurement leaders, and open-source maintainers
- Technologists and security teams: Prioritise asset inventories, rapid patching, strong authentication and reducing unnecessary internet exposure before deploying experimental AI tools. Use AI where it fills a clear gap in capability and ensure human expertise validates outputs.
- Procurement leaders and boards: Judge cybersecurity investments by measurable risk reduction rather than by the amount of defensive AI in a product. Before approving an AI security purchase, ask what specific risk it reduces and whether a cheaper, more predictable control would reduce that risk better.
- Open-source maintainers: Prepare for and mitigate AI‑generated noise. The ASD highlights already‑reported floods of bogus AI-generated bug reports as a real operational problem that can consume scarce attention and obscure genuine issues.
Questions vendors must answer
The ASD recommends that organisations ask hard questions of vendors pitching “AI‑powered” security tools. The agency suggests asking:
- What exactly does the AI component do, and what would the product do without it?
- How often are its findings incorrect, and how is that measured?
- What data leaves the network, and where does it go?
- What permissions does the product need, and what can it do without human approval?
- How are models updated and tested, and for how long will the vendor support them?
According to the ASD, a vendor unable to answer those questions clearly is “selling a label, not a capability.”
Measured against the ASD’s findings, the practical test for boards and executives is simple: start with the organisation’s biggest risks and choose controls that demonstrably reduce them. AI can help in particular contexts, the report says, but success is ultimately measured by how much risk is reduced — not by how much AI is being used.
