Client Portal
What ASD published
The publication is Frontier AI cyber threat considerations for boards of directors, produced by the Australian Signals Directorate together with the Australian Institute of Company Directors. It is short, direct, and unusual: ASD normally writes for the people who run systems, and this is written for the people who own the risk.
Its argument is that frontier AI models, the current generation of advanced AI systems, are changing what attacks cost and how fast they move. It gives boards sixteen threshold questions to put to management, and a prioritised list of what to do about the answers, starting with securing attack surfaces and patching known vulnerabilities and ending with Secure by Design principles for new systems.
What actually changes when attackers use AI?
Three shifts, all taken directly from the ASD publication.
- Speed. ASD says AI can cut vulnerability discovery and exploitation timelines from days to hours. An attack window that used to leave time for a patch cycle may not any more.
- Chaining. AI can combine several low-severity weaknesses into one high-impact compromise. Problems that were individually not worth an attacker's time become worth an algorithm's time.
- A lower barrier. People who previously lacked the knowledge or skill to attack anyone now have capable tools, including through abuse of AI model guardrails. ASD's phrase is that this amplifies the traditional asymmetry between attackers and defenders.
The third shift matters most at small-business scale. The comfort of being too small to bother with was always thin. It does not survive attackers whose marginal cost of one more target is close to zero.
Who asks the questions when there is no board?
ASD's sixteen questions assume a boardroom: directors on one side, management on the other, and someone whose job it is to have the answers. At 5 to 50 staff you are both sides of that table. That does not make the questions irrelevant. It makes them harder to dodge, because there is nobody else they can be delegated to.
Here are the five we would put at the top for a cloud-first business running on Microsoft 365, with honest answers.
Could someone with no technical skill attack us now?
Yes. That is the core of ASD's warning. The consequence is volume: more attempts, better written, more convincingly targeted. The defence does not change, it just stops being optional. Phishing-resistant multi-factor authentication and patched, supported systems are named in ASD's own priority list.
Could several small weaknesses become one large incident?
This is ASD's chaining point translated. A leaver's account that was never disabled, a mailbox without MFA, a SharePoint site shared more widely than anyone remembers: each is minor on its own, and together they are a data breach. One of ASD's questions is whether you are delaying fixes because each one looks low-severity. Most businesses are. The list of things you meant to tidy up is exactly what chaining exploits.
Which AI tools are we relying on, and did anyone assess them?
ASD asks boards to weigh their reliance on AI vendors, including cyber supply chain risk and foreign ownership, control or influence. The small-business version is more basic: do you know which AI tools your staff are using at all, and what business data is going into them? If the answer is no, start with an approved list and an acceptable use policy, which we have covered in governing your team's AI use.
If an attack took hours instead of days, would we notice?
ASD asks whether detection and response would remain effective if attacks moved at machine speed. The honest answer for a business with no monitoring is no, because alerts nobody reads are the same as no alerts. Detection is not a product you switch on. Someone has to be watching, and accountable for watching.
Could we keep trading through an incident?
The guidance closes on continuity: could your most important services keep operating if a critical system was compromised, and have your response and continuity plans been tested rather than merely written? For most small businesses the test is simpler. When did anyone last restore a file from backup to prove the backups work, and does a plan exist on one page that says who to call? We have written a first-hour guide worth keeping printed.
The defence is less exciting than the threat
Here is the part of the ASD publication that deserves the most attention and will get the least. Its recommended response to a frontier AI threat is not frontier AI defence. The immediate priorities are securely configured systems, patched vulnerabilities, strong identity and access controls, minimum privileges, a plan for legacy systems, and tested incident response. ASD also asks whether you adhere to a recognised cyber security framework.
None of that is new, and that is the point. The threat got faster; the defence got more urgent, not different. For Australian small business the recognised framework is SMB1001, and it is the standard we align our own managed security work to.
AI defends as well as it attacks
The same guidance is not a warning against AI. Among its medium-term priorities is adopting AI for cyber defence, kept secure, human-supervised and accountable, and ASD publishes companion guidance on exactly that, titled Opportunities for AI in cyber defence. AI is already better at reading logs at volume than any human, and the same speed that helps attackers helps the people watching your environment. The asymmetry ASD describes is only fatal for organisations where nobody is watching at all.
What to do with this in one month
- Ask the five questions above and write the answers down. An honest page of answers is worth more than a shelf of policies.
- Run the Cyber Health Check. It is free, takes about three minutes, and scores you against SMB1001, the framework question ASD says someone should be asking.
- Fix the chaining candidates first. Old accounts, mailboxes without MFA, and sharing nobody remembers approving. They are cheap to fix and they are what an automated attacker will find.
If the answers worry you, that is what a conversation is for. No obligation, and we will tell you if you are in better shape than you think.