Frontier AI Is Accelerating Cyber Attacks—What Happens Next Depends on Resilience

When news broke that unauthorised users had accessed Anthropic’s Mythos model, my phone lit up. I saw it in real time, but my reaction wasn’t panic; it was recognition.

Most of us in the security industry have believed for some time that this moment was coming. Whether it arrived via unauthorised access, as happened, or through frontier AI (Artficial Intelligence) models gradually entering the public domain, the trajectory was never in question. The question was only when, and as it turns out, the answer is now.

Frontier AI Has Made This a Speed Problem, and We’re Losing Ground

Let me be direct about what frontier AI models change for attackers. It’s changed everything, and most of all, the clock.

Attackers have always been able to find vulnerabilities and write exploits. We can debate the relative effectiveness of human versus machine capability, but nobody is debating whether this is all going to happen faster. It will, and dramatically so.

The problem is that our defences have also been powered at human speed, human brains, human teams, and human processes. There is an asymmetry building in real time that the security industry across Asia Pacific is not yet equipped to handle.

Here’s the second-order problem that keeps me up at night. It’s not just that attackers find the same vulnerabilities faster. Frontier AI models will uncover more vulnerabilities that humans ever could; faster and exploited at machine speed. The math is not on our side.

Consider patching, one of the most persistent failures in our industry’s history. We’ve spent decades debating why organisations struggle to patch within acceptable timeframes. In a world of machine-speed attacks, those timeframes collapse. We are not going to patch fast enough to keep up.

That matters because much of the software ecosystem relies on open-source components maintained by communities with limited time and resources. Attackers with access to frontier AI models may identify and weaponise vulnerabilities faster than those communities can realistically respond.

The challenge is not simply moving faster. The challenge is that we do not yet have an operating model designed for this reality.

The Math Demands Cyber Resilience

If frontier AI enables attackers to identify vulnerabilities more quickly and launch exploits at machine speed, the math becomes difficult to ignore. The likely result is more breaches. Not because defenders are failing, but because the scale and speed of AI-driven attacks make perfect prevention increasingly unrealistic.

This is why breach containment must sit at the centre of how organisations across Australia and Asia Pacific rethink their security model. What matters is what happens next: how effectively you contain inevitable breaches, how you continue operating while under attack, and how quickly you recover. Cyber resilience isn’t a secondary priority to defer until prevention is solved; it is the strategy.

Regulators in our region are already reflecting this shift. Australia’s Critical Infrastructure Risk Management Program and the Protective Security Policy Framework are pushing organisations toward operational resilience as a baseline requirement, while Singapore’s Cyber Security Agency has made resilience central to its national strategy.

Australia’s Cyber Security Centre has also warned about the impact frontier AI models could have on cyber threats. Its guidance specifically recommends applying isolation controls such as network segmentation and segregation to limit pathways for compromise. It also highlights the need to responsibly apply AI in defensive operations to improve existing manual processes and help organisations respond faster at scale.

This reflects where the threat environment is heading and how quickly it is evolving.

For industries that run on uptime, from financial services and healthcare to energy and logistics, the calculus is straightforward: assume breach, architecturally contain it, and recover fast.

Speed and cyber resilience are the two things defenders must get dramatically better at, and two areas where the industry hasn’t historically excelled. That must change.

Stop Watching the Frontier AI Model Leaderboard

I regularly speak with enterprises across the region about AI, and the conversation tends to follow the same pattern: which model, which version, which capability?

Right now, Mythos is driving the conversation. In two months, it will likely be a different frontier AI model. The underlying dynamic, however, won’t change.

China is making a sustained push on its own frontier models, and there is no reason to expect that pace to slow. The open-source versions of today’s most advanced capabilities will become widely available—it’s a question of when, not if.

Distillation isn’t a risk to plan around; it’s a base assumption. For organisations operating across Asia Pacific, this has real strategic implications.

What I tell every customer is this: don’t build a defence strategy around which frontier AI model is leading the threat landscape this week. Build an operating model for a world of models, one where the capabilities that concern us today will be cheaper, more accessible, and more widely distributed than they are right now. The time between a model’s release and its open-source equivalent is compressing, and it will continue to do so.

We Have a Window, Don’t Waste It

What isn’t discussed often enough is that we’ve had clear warning signs. The trajectory of frontier AI models has been visible for some time, and Mythos did not emerge in isolation. There were developments leading up to this moment, as there often are before major technological inflection points.

The fact that Anthropic limited the release of a model because of cybersecurity concerns matters. It signals that frontier AI capabilities have crossed into territory where the risks are no longer theoretical. That should change how organisations think about the threat landscape.

The priority now is not trying to patch faster than machine-speed attacks. The focus needs to shift towards breach containment, operational agility, and resilience by design.

Organisations need visibility into exposure paths, trust relationships, and how attackers move through hybrid cloud environments. They need containment-focused exercises, faster crisis response, and architectures designed to limit lateral movement before an incident occurs. They need cyber resilience.

The threat environment organisations are planning for today already risks looking outdated. The warning signs are here. The window to adapt is open, and so is the ability to improve cyber resilience. What organisations cannot afford to do now is waste it.