Supercharged security: Cyber risk in the age of frontier AI

HKCERT report reveals that cyberattacks have become more automated, targeted, and destructive with the rapid proliferation of AI, posing significant threats to business operations and information security. A record high 15,877 cybersecurity incidents were recorded in Hong Kong in 2025, marking a 27% year‑on‑year increase and underscoring the accelerating pace of cyber risk.

At the core of this shift is AI's ability to compress the timeline of cyber risk. The gap between vulnerability disclosure and exploitation is collapsing, with weaknesses that once took weeks or months to identify now exploited in hours. This acceleration is fundamentally changing how organizations must approach security.

The shift cuts both ways. The same models accelerating attackers are also enabling defenders to detect, analyze, and mitigate risk at unprecedented speed. The advantage is no longer about better tools, but about operating faster and more cohesively across the security lifecycle. Organisations relying on slow, fragmented approaches will struggle to keep pace. Security must now function as a continuous, integrated process spanning development, detection, and response.

Traditional Security Tooling Meets Its Limits

Product security testing has remained largely unchanged for years. Secure development life cycles typically include static analysis, software composition analysis, dynamic testing, fuzzing, penetration testing, and manual code audits. All of these are necessary for meaningful coverage, often requiring multiple tools in each category.

However, traditional security tooling - with known limitations and manual dependencies - is reaching an inflection point. Frontier AI models are overtaking static approaches in their ability to analyze complex systems quickly and holistically. These models can read and reason across entire codebases and memory, simulate workflows by agent, and identify weaknesses at speeds previously unattainable by human‑driven processes alone.

Frontier Models and the Changing Threat Landscape

One of AI's most significant impacts is the democratization of development. People outside traditional engineering roles can now generate functional applications in short timeframes. From a security perspective, this capability cuts in two directions: the same models that help detect and remediate vulnerabilities can also accelerate their exploitation.

Recent attention around frontier AI initiatives, including early access programs for critical software maintainers, reflects a broader industry shift. These models are rapidly advancing in their ability to analyze, understand, and test complex systems. The implication for defenders is clear: staying ahead of threat actors now depends on adapting as quickly as the tools themselves evolve.

Security at Breakneck Speed

Frontier AI models increasingly resemble all‑purpose security experts - combining the capabilities of seasoned penetration testers and code auditors. They can dramatically accelerate vulnerability discovery, completing in hours what once took weeks. They can assist in identifying zero‑day vulnerabilities and subtle systemic weaknesses that evade conventional testing.

Yet these models are not infallible. False positives still occur, and identified issues may already be mitigated elsewhere in the system. Human-in-the-loop analysts remain essential to guide models, validate findings, and make informed decisions. What changes is productivity: analysts can move faster, focus on higher‑value work, and reduce exposure windows through earlier intervention