AI Wales

Following the success of last year's Cybersecurity in AI meetup, we've teamed-up again with Matt Hopkins to bring you an updated event for 2025.

We'll have two cybersecurity experts giving a talk each, and there'll be time for some Q&A as well. Space is limited for this event, so secure your spot today - but do definitely attend if you take a space.

Doors will open from 18:30, and we will begin promptly at 18:45. We aim to finish by 20:45, for those catching the train or have other commitments.

Talk 1: AI's Blind Spots: Bias, Misdirection, and the Human Solution
This talk explores how AI can be misled or biased, using real examples. We’ll use tools used in industry to reveal what models are actually seeing and we’ll highlight biases from companies like Amazon or Apple. We’ll also discuss the role of the "human in the loop" as a solution to improve AI fairness and reliability. Finally, we’ll open a conversation about the limits of machine understanding.

About Hamza Afzal
As a researcher at Cardiff University, Hamza looked at the advantages and disadvantages of AI for the UK's road network - like using AI road cameras on the M4 to automate traffic violations - and their UK-specific implications.

Funding from National Highways facilitated the year’s study, resulting in a paper awaiting publication, and a pilot project involving a Raspberry Pi, a toy car, and a suitcase. With this kitchen-sink model of an "adversarial attack", Hamza and his colleague demonstrated how
something like a National Speed Limit sign could influence AI-powered technology and cause large-scale damage to the county’s network.

Post-research, Hamza & his colleague Bilal saw an opportunity to join the University’s entrepreneurship programme with Dr. Eirini Anthi - the research project’s co-supervisor. The 3 co-founders built an AI-system measurement tool. Hamza was able to question both academia and industry, giving him a uniquely lateral look at AI systems as a whole.

Talk 2: AI Security in Action
In this talk, John will navigate the evolving landscape of AI security guidance—from OWASP and NIST to MITRE, NCSC, and beyond. He’ll explore the real-world challenges that teams face today, from information overload to unclear responsibilities, and how the UK AI Cybersecurity Code of Practice (now formalised as an ETSI standard) can help simplify and accelerate adoption. Drawing on practical experience, John will make the case for collaboration and community as the most effective antidotes to complexity fatigue—bringing AI security from theory into action.

About John Sotiropoulos
John Sotiropoulos is the head of AI Security at Kainos where he is responsible for AI security and securing national-scale systems in government, regulators, and healthcare. A Board Director for OWASP GenAI Security and a co-lead for the OWASP Top 10 for LLM Applications, John also leads the Agentic Security Initiative. At OWASP he is responsible for alignment with other standards organizations and national cybersecurity agencies including NIST, MITRE, CSA, the NCSC, CISA, and the US AISIC, where he is the OWASP lead.

He is the author of the bestselling book "Adversarial AI, Attacks, Mitigations, and Defense Strategies" and more recently he authored the official UK Government’s Implementation Guide for the UK Code of Practice for the Cybersecurity of AI, both now international ETSI standards.