Data Integrity, AI Risk, Cyber Realities and tech leadership with Kate Carruthers скачать в хорошем качестве

Data Integrity, AI Risk, Cyber Realities and tech leadership with Kate Carruthers

In this episode of the Technology & Security podcast, host Dr. Miah Hammond-Errey is joined by Kate Carruthers. Kate is currently the head of data analytics and AI at the Australian Institute of Company Directors. She shares her journey from defending Westfield against state and non-state cyber attacks to leading UNSW's enterprise data, AI, and cybersecurity efforts, including delivering the university's first production AI system in 2019 and re-architecting its cloud data platform for AI and ML. She notes boardrooms are evolving from basic cyber literacy to probing AI risks like models, data, and risk registers.    Carruthers outlines some real-world examples, such as UNSW’s enterprise AI program, including a machine learning model that predicted which students were likely to fail a course, with 95%+ accuracy, so the university could design careful, humane intervention protocols to reduce self-harm risk. She argues that while frontier models like OpenAI and Gemini have a place, their compute costs, water intensity and general-purpose design make them poorly suited to some business problems, and that the future lies in smaller, industry-specific models trained on highly relevant data. The conversation covers the rise of agentic AI coding tools, the risk of deskilling junior developers, and the need for diverse, product-focused teams to translate technical systems into workable human processes.​   On security, she prioritizes CIA triad integrity over confidentiality, warning of data alterations in cars, medical devices, and government systems via poisoning or underinvestment in encryption. Carruthers urges Australian AI sovereignty—opting for open-source like Databricks over proprietary stacks—amid US-China model contrasts and outage risks from providers like AWS or CrowdStrike. Throughout, she encourages leaders not just to read about AI but to use multiple systems themselves, understand their limitations as probabilistic tools in deterministic business environments, and ground every deployment in clearly defined problems, ethics, and user needs.​