With Anupam Mediratta (Stealth Startup) – Building Adaptive IoT Security with AI-Generated Attack Simulation and Florian Lukavsky (SignPath) with all the CRA latest.

IoT attacks have surged 400% year-over-year, with AI-powered threats now bypassing 78% of traditional security measures.

Manufacturing faces 906 unique threat signatures, while critical infrastructure absorbed 71 billion attack attempts in 2024 alone.

Traditional detection methods are fundamentally broken.

This talk presents groundbreaking research that flips the script on IoT security: using AI to generate synthetic “hard-negative” attacks that train smarter, more resilient detection systems.

We combine diffusion-based generative models with pre-trained time-series foundation models to create attacks that mirror real-world APT tactics—from polymorphic malware to living-off-the-land techniques.

Our approach specifically addresses the critical gap in current IoT security: the insufficient diversity of attacks in training data.

By generating synthetic attacks that maintain benign statistical properties while executing malicious actions, we prepare defences for threats that don’t yet exist in the wild.

With malicious APT groups targeting cyber-physical systems and “malware sentience” emerging through AI-powered autonomous attacks, this research offers a critical path forward.

We’re not just detecting known threats—we’re anticipating and defending against the AI-enhanced attacks of tomorrow.

Anupam Mediratta bio: Anupam is an entrepreneur in the field of AI and Machine Learning. He was previously a PhD student (now a dropout) at the University of Edinburgh, where he became interested in applying AI to IoT security. He has been an IoTSF professional member for the last few years, which has helped him understand the space of IoT Security. He has also been a tech/ML lead at Microsoft, Uber and a couple of startups. He also holds a few papers/patents in the field of AI.