How AI 'Model Harnesses' Are Changing the Cyber Threat Landscape

The Australian Cyber Security Centre (ACSC) has published an update on frontier AI models and the software frameworks—known as 'model harnesses'—that connect these AI systems to external tools, data, and networks. These harnesses allow AI models to take actions beyond simply generating text, such as running code, browsing systems, or interacting with other software, which significantly expands what AI can do in both offensive and defensive cyber security contexts.

As AI models become more capable, the harnesses built around them are becoming a critical factor in determining real-world impact. The same technology that helps defenders automate detection and response could potentially be misused to scale up attacks, automate reconnaissance, or assist less skilled attackers in carrying out more sophisticated operations. The ACSC's update is intended to help organisations understand this evolving capability so they can better anticipate how AI may influence the threat environment.

For Australian small and medium businesses, this development is a reminder that AI is becoming deeply embedded in the tools used by both attackers and defenders. Businesses adopting AI-powered products should understand what these systems are capable of and how they are secured, while also watching for signs that AI is lowering the barrier for attackers to target smaller organisations.