AI Self Defence

AI Self Defence

AI is now built into the tools we use every day, and the risks are changing faster than traditional security thinking can keep up. This talk is a practical guide to the security, privacy and operational risks that come with modern AI.

We’ll look at how AI systems actually fail and how bad actors exploit them, using recent real-world cases such as the Hugging Face supply-chain breach. You’ll come away understanding the main threats, including prompt injection, data leakage and automated social engineering.
Then we’ll get practical. You’ll leave with “self-defence” strategies you can use straight away:

• A simple framework for working out your own personal threat model
• How to keep sensitive data out of model training
• What to look for when evaluating a vendor’s guardrails
• Habits that keep you resilient as the tools keep changing

Whether you’re a developer, a professional, or just curious about tech, this session will help you use AI confidently and safely.

About the speaker

Emma Humphrey is the Director of AI Safety New Zealand. Originally from Christchurch, with a background in physics and mathematics, she spent a decade overseas in data science startups before joining the UK Home Office. There she spent four years leading high-stakes projects in highly secure environments: designing tools to detect deepfakes and criminal exploitation, building digital tools to support sexual assault survivors, and auditing automated hiring systems for hidden bias.

Since returning home, Emma has focused on the non-profit and research space. She organised New Zealand’s inaugural AI Safety Conference and launched the country’s first AI Safety co-working space for researchers and non-profits. Through the Nelson AI Sandbox, she has helped more than 150 non-profits across Christchurch adopt AI safely and save time with the right tools.