When Attacks Move at Machine Speed: What Boards Need to Ask About Frontier AI

Most cyber risk assessments carry a quiet assumption: that attackers need time. Time to find a weakness, time to work out how to use it, and time to string small problems together into something serious.

Frontier AI is dismantling that assumption, and on 5 August 2026 the Australian Signals Directorate, working with the Australian Institute of Company Directors, put it in writing for boards.

The ASD publication, Frontier AI cyber threat considerations for boards of directors, is direct about what has changed. Frontier AI has, in the ASD’s words, the potential to “significantly reduce vulnerability discovery and exploitation timelines from days to hours while significantly lowering the skill and knowledge barrier for malicious actors”.

Days to hours. And the people capable of it are no longer a small, highly skilled group.

What has actually changed

The ASD describes frontier AI models as those with advanced reasoning, software development, natural language or multimodal capabilities, trained at scale, and able to perform complex tasks beyond the reach of conventional automated tools. For cyber security, three capabilities matter most:

• Identifying vulnerabilities and rapidly weaponising them.
• Chaining multiple low severity vulnerabilities into high impact compromises.
• Performing malicious cyber activity with little to no human oversight.

That second point deserves attention. Plenty of organisations carry a backlog of low and medium findings that never reach the top of the queue. The ASD warning is that machine speed analysis can combine those individually unremarkable weaknesses into a genuine incident. Low severity is not the same as low risk when something can test thousands of combinations in an afternoon.

This is not theoretical

In November 2025, Anthropic published details of what it described as the first reported AI orchestrated cyber espionage campaign, in which attackers used agentic AI to automate the bulk of an intrusion campaign against technology companies, financial institutions and government agencies, with human operators intervening at only a handful of decision points. The campaign has since been catalogued by MITRE ATT&CK.

Google’s Threat Intelligence Group reports a similar shift, finding that adversaries are no longer using AI just for productivity gains and are now deploying novel AI enabled malware in active operations.

The baseline this lands on is already under strain. ASD’s Annual Cyber Threat Report 2024 to 2025 recorded more than 84,700 cybercrime reports, roughly one every six minutes, with the average self reported cost per report rising to $56,600 for small business, $97,200 for medium business and $202,700 for large organisations. Now add speed and scale to that picture.

“Boards do not need to understand how these models work. They need to be confident management can show, with evidence, that the organisation could still operate if an attack unfolded in hours rather than days.” Robert Buck, Managing Director, Diamond IT

The questions your board should be asking

The strength of the ASD and AICD guidance is that it gives directors a starting point that does not require technical fluency. Among the threshold questions it recommends boards put to management:

• How vulnerable are we to AI enabled attacks, and what assumptions in our current risk assessments might frontier AI invalidate? Is our risk tolerance still appropriate?
• If frontier AI were used to find and exploit weaknesses across the organisation, where would we be most exposed? Where could minor weaknesses in our systems and cyber supply chain be combined into a major incident?
• Are we relying on vendors without sufficient governance and oversight, including an understanding of their foreign ownership, control or influence? Do we have visibility of third and fourth party suppliers?
• Do we have control of the fundamentals, such as adherence to a recognised cyber security framework? What legacy technology risk are we carrying, and are we delaying fixes because the exposure looks low?
• If attacks moved from taking days to hours, could we still detect and respond effectively, and keep our most important services running? Have our incident response and business continuity plans been tested for this?

If those questions produce confident, evidenced answers from management, that is a good sign. If they produce hesitation, that is the finding.

The response is mostly disciplined basics

Here is the genuinely reassuring part. The guidance does not call for a new category of AI defence spending. It calls for doing the fundamentals properly, and sooner, across four horizons.

Immediately, secure your attack surfaces through approved, maintained configuration baselines with drift detected and corrected, and reduce software vulnerabilities through continuous identification and risk based remediation timeframes.

In the short term, manage or replace legacy systems that cannot meet security requirements, reinforce identity, credential and access management including phishing resistant multi factor authentication, restrict unnecessary privileges so that people and AI agents hold only the access their duties require, and make sure incident response and business continuity plans are current and genuinely exercised.

In the medium term, adopt AI for cyber defence in a way that is, in the ASD’s words, “secure, controllable, human-supervised and used in an ethical and accountable manner”. Defenders get to use these capabilities too, for vulnerability identification, security assessment and triaging security events. Longer term, modernise using Secure by Design and Secure by Default principles.

If that list feels familiar, it should. It is the Essential Eight mindset, identity first security and tested resilience, work that many organisations have had on the plan for a while. What has changed is the cost of deferring it. Organisations that fail to act now, the ASD warns, may be forced to accept greater business risk and exposure to losses.

Do not forget the AI you are bringing in the front door

The guidance makes a point that is easy to skip past. The risk is not only about what attackers do with AI. It is also about the AI your organisation is adopting. Boards are asked to consider cyber supply chain risk tied to reliance on particular AI vendors, and the foreign ownership, control or influence risks that may come with it.

Every AI tool, agent and integration added to your environment is a new identity, a new set of permissions and a new dependency. The ASD principle of least privilege explicitly extends to AI agents. If you are deploying agents that can read email, access files and act across systems, what they are permitted to do, and who reviews that, belongs on the risk register alongside your human access reviews.

Where to start

You do not need a frontier AI strategy to respond to this. You need three things on the next board agenda:

1. A reassessment of risk tolerance against a threat environment that now operates at machine speed rather than human speed.
2. An honest view of the fundamentals, covering configuration baselines, patching timeframes, identity and access controls, privilege management, legacy systems and cyber supply chain visibility, supported by evidence rather than assurances.
3. A tested response capability, with incident response, business continuity and disaster recovery plans reviewed and exercised for faster, more automated attacks.

The ASD closing message to directors is unusually plain: boards should be pressing management to act now, including by supporting targeted investments to enhance cyber security and resilience where appropriate.

How Diamond IT can help

We work with boards and executive teams to translate guidance like this into a practical plan, from Essential Eight uplift and identity hardening through to legacy system remediation, incident response planning and the secure adoption of AI and Microsoft Copilot. If your board is asking the questions above and you would like help answering them with evidence, get in touch with our team on 1300 307 907, send an email to enquiries@diamondit.com.au or fill out the form below.