Threat Intelligence
ACSC Threat Report 2025-26: The AI Threats
26 Oct 2026 · Updated 26 Oct 2026
The 2025-26 report makes AI its central theme, with a clear framing: AI has amplified existing cyber threats rather than created new ones. It lets attackers find vulnerabilities and build exploits faster, sharpens social engineering through convincing impersonation, and, with agentic AI emerging, may accelerate the pace and scale of attacks further.
Has AI created new kinds of attack?
No, and that framing matters. ASD’s assessment is that AI has not fundamentally changed how adversaries gain access and move through an environment, and has not introduced new cyber threats, but it has amplified existing ones, making attacks faster, cheaper and available to more people. That is a more useful way to think about AI risk than treating it as a wholly new danger: the defences that worked still work, but the assumptions about attacker cost and skill no longer hold. This is the same logic behind the new ISO/IEC 27090 guidance and AI considerations in the ASD Essentials, both of which add AI specific thinking on top of established security.
How are attackers using AI?
To move faster through the attack chain and deceive more convincingly. ASD reports that both state actors and cybercriminals can use AI to run reconnaissance, identify vulnerabilities and develop exploits or malware more quickly, and to launch attacks at greater scale. AI is reducing the expertise and effort needed to find and exploit vulnerabilities, which ASD says likely accelerates the time to exploit and widens the pool of capable actors. On the human side, AI has sharpened social engineering: synthetic voice and video can impersonate trusted people to deceive help desks or those who approve payments or access, a risk covered in deepfake and voice-clone fraud. Faster exploitation makes prompt patching more important than ever.
Can AI systems themselves be attacked?
Yes, and the report and the new standards both address it. Beyond using AI to attack, adversaries target AI systems directly. ASD points to risks such as indirect prompt injection through emails, documents or web content processed by AI agents with access to tools or data, tampered models pulled from public sources, and sensitive data sent to unapproved AI services. These are exactly the AI specific compromises catalogued in ISO 27090’s threats and mitigations, covering data poisoning, model theft and more. For organisations deploying AI, securing the AI itself is now part of the job, not an afterthought.
What is the concern with agentic AI?
Autonomy could accelerate the threat further. ASD specifically flags the emergence of agentic AI, systems that act with autonomy to carry out tasks, and the potential for rapid advances toward highly autonomous or recursively self-improving AI, as factors that may accelerate the pace, scale and sophistication of cyber threats and require ongoing investment. ASD also notes new threats emerging from the unintended actions of AI agents and model testing. The caution is not hypothetical: as organisations give AI agents identities, permissions and access, those agents become part of the attack surface, which is why the ASD Essentials treats them as assets to govern like any other identity.
Can AI help defenders too?
Yes, with governance. The report is balanced: AI-enabled security tools can process large volumes of data, correlate events, identify vulnerabilities, triage alerts and synthesise threat intelligence, helping defenders respond faster. The condition is that these tools are themselves AI assets to be governed, and that people remain accountable for decisions and validate AI outputs. ASD’s position is to use AI to support good security practice, not replace it. For organisations building AI governance to do this well, Cybernion helps through AI governance and ISO 42001 readiness.
Frequently asked questions
Has AI created new cyber threats?
ASD’s assessment is no. AI has amplified existing threats rather than introducing new ones, making attacks faster, cheaper and available to more actors, which makes action more urgent.
How is AI used in attacks?
To run reconnaissance, find vulnerabilities and build exploits more quickly, and to make social engineering more convincing through synthetic voice and video impersonation.
What is agentic AI and why does ASD flag it?
Agentic AI acts with autonomy to carry out tasks. ASD warns that agentic and highly autonomous AI could accelerate the pace, scale and sophistication of cyber threats, needing ongoing investment.
Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.
If AI is changing your risk and you want help, talk to us.
Sources:
- Annual Cyber Threat Report 2025-26, Executive summary, cyber.gov.au, 26 October 2026
- AI considerations within the Essentials, cyber.gov.au, 15 October 2026
- ISO/IEC 27090:2026, iso.org, October 2026
Last updated: 26 October, 2026
