Australia’s cyber alerts rose sharply. Its AI attack count is unknown

The Australian Signals Directorate (ASD) says in its 2025–26 Annual Cyber Threat Report, published 26 October, that it issued more than 4,400 warnings, up almost 80%. The report gives no AI-attributed attack count.

The big change

• What changed: ASD’s near-80% rise measures warnings; incident responses rose about 20%.
• Why it matters: Verify sensitive requests through a trusted channel; convincing words or voices are weak proof of identity.
• What to watch: An AI-attributed incident series and a disclosed case definition would make claims about AI’s share testable.

The report invites two different readings. One concerns how often its cyber security centre saw or acted on a problem. The other concerns what new tools might make an attacker more effective. Treating a change in the first as proof of the second would assign a cause that the published evidence does not establish.

Reports, responses and warnings

ASD received over 82,000 cybercrime reports and handled over 1,500 incidents, up 20%. Its notes describe distinct, incomplete datasets.

These measures have different units. A report, an incident response and a warning are not mutually convertible. Comparing their changes requires knowing how reporting, detection and outreach worked in each period. The headline totals alone cannot identify which of those processes changed.

The 2024–25 ASD report recorded more than 84,700 cybercrime reports, more than 1,200 incident responses and more than 1,700 notifications. Each figure is a rounded lower bound. Dividing those bounds would manufacture a precise growth rate.

Even comparisons within the incident series need care. The current year-in-review flags revised severity categories and improved reporting. The count can change because more activity happened, more was detected or more was reported. These explanations cannot be separated using the headline totals alone.

AI changes the approach to a target

ASD’s social-engineering chapter describes AI-written phishing; generative tools can produce plausible messages quickly and at scale, reducing the skill and effort required.

Australia’s corporate regulator provides a more specific example from its own remit. ASIC says it has seen investment scams built around deepfake videos of public figures. The videos sit alongside spoof websites, fabricated articles and fake reviews. A search for the supposed opportunity may therefore turn up several mutually reinforcing false signals. ASIC reports removing more than 19,400 online scams in the financial year, up 182%, but its removals include many kinds of fraud and reflect enforcement activity. They cannot be used as a count of AI attacks.

The UK’s National Cyber Security Centre annual review describes a similar pattern across a different jurisdiction: threat actors use AI to improve existing techniques, including reconnaissance, social engineering and vulnerability research. It expects AI-assisted vulnerability research and exploit development to matter more, which is an assessment of likely future capability. The UK account corroborates the mechanism. It does not independently validate a percentage change in Australia’s incidents.

What a warning can and cannot tell its recipient

An ASD warning prompts investigation. The recipient may still be able to check exposure, identify a compromise or contain one. Without the finding from that investigation, the warning alone should not be described as a successful intrusion. Preventing an intrusion is also a reason to warn early, so the value of an alert cannot be judged only by the number of incidents later confirmed.

ASD’s separate social-engineering guidance advises independently verifying unexpected requests through a trusted method, especially when they ask for credentials, a payment or a security-setting change. That shifts the check away from the appearance of a message or the sound of a caller. ASIC makes a parallel point for investments: polished search results and familiar endorsements are not adequate verification of a provider. These are concrete precautions supported by the agencies’ guidance; this article has not tested their effectiveness in a particular case.

To estimate AI’s share of Australian cyber incidents, readers would need a consistent definition of AI involvement, case-level attribution and a disclosed denominator over comparable periods. That denominator is the missing evidence behind any claim that AI drove the increase.