Cyberattack on Taiwan Government Signals a New Era of Autonomous Threats

What happened in Taiwan matters far beyond this individual attack. We are crossing a fundamental boundary in cybersecurity: from humans using machines to execute attacks to autonomous machines capable of pursuing an objective.

We have had machine-speed cyberattacks for decades. DDoS and botnets gave attackers enormous scale. One human could control thousands or millions of machines and point enormous computational force at a target.

But those machines were largely brawn. The intelligence remained with the human.

Humans acquired or compromised the infrastructure. Humans determined the attack geometry. Humans chose targets and tactics. And when defenders changed the environment, humans had to observe what happened, rethink the attack and redirect the machines.

That distinction matters because traditional automation is predictable. If I understand what the machine is doing, I can build defenses against that behavior. In a DDoS attack, for example, I can filter, rate-limit, scrub or, in extreme circumstances, blackhole traffic. I am defending against machines executing a pattern.

Agentic AI changes that equation because intelligence is now inside the loop.

An autonomous agent doesn't simply execute faster. It senses what happened. It reasons about why it happened. It changes its approach and tries again.

Block one path and it can search for another. Change the environment and it can change with it. Deploy a defense and the defense itself becomes new information that the attacker can reason about. That is a fundamentally different adversary.

The important word isn't simply intelligence. It is adaptation.

This is why I think of this as a new form of intelligent DDoS — not DDoS in the traditional network sense, but a DDoS of cognition and decision-making.

Traditional DDoS distributed computational force. Agentic AI can distribute adaptive intelligence.

Instead of a million machines repeatedly doing what one human told them to do, imagine a million agents capable of observing their individual environment, reasoning about what they encounter, changing tactics and continuing toward an objective.

Now the attack geometry itself becomes dynamic.

That changes the economics of cyber warfare. Expertise becomes software. Intelligence becomes reproducible. Adaptation becomes automated. And sophisticated attack capability that historically required scarce human talent can potentially be instantiated and operated at enormous scale.

It also exposes the fundamental weakness of today's defensive architecture. You cannot defend against machine-speed adaptation with human-speed decision making.

A SOC ticket, an analyst investigation, an escalation and a change-control meeting are not a closed loop capable of competing with an adversary that is sensing, reasoning, acting and learning continuously.

I have said for some time that the only way to stop AI is with AI.

Cyber defense itself has to become an adaptive autonomous system: sense → reason → act → validate → sense again. Humans move above the execution loop, defining objectives, constraints, permissions and rules of engagement.

We are moving from humans fighting humans with machines, to machines fighting machines under human governance. And the competitive advantage will increasingly belong to the system with the fastest trusted adaptive loop.