Florian Oelmaier
Expert for IT crisis management, digital forensics and the aftermath of ransomware attacks, @h0tz3npl0tz
Florian Oelmaier is CTO of IS4IT and an expert in current attack vectors on computer applications and networks (IT, OT and ET). His motivation is Digital Sovereignty. Mr. Oelmaier is member of the Expert Advisory Board Cyber Security for the German Federal Office for Information Security. He has published numerous articles in professional journals, multiple books (the last “Crisis Case Ransomware”) and is one of the authors of the “NSA report”, analyzing the actions of state sponsored actors in Europe. He loves giving keynotes and talks at conferences and events.
TAlks & Q&A
Free the Beast from its Guardrails: Open Weights AI - Digital Sovereignty or Chinese Trojan Horse?
Description:
While the West holds panels about AI regulation, China ships models. GLM-5.3-flash and DeepSeek-V4.1-flash run on your own hardware, cost a fraction of the US top tier – and promise exactly what Europe has been longing for: digital sovereignty through open weights. Open — or is it? In this talk, we take an honest look at the real quality gap between Chinese open-weights models and the American top tier: Where do they match, where do they fall apart, and why the benchmark numbers lie to you.
We dissect what these models actually „think": how chain-of-thought reasoning works in GLM-5.3-flash and DeepSeek-V4.1-flash, what they learned from the Vibethinker generation, and why hybrid thinking models at flash speed are the true game changer for sovereign and edge AI. Then we go where vendor keynotes never go: inside the guardrails. How are they architected into these models, what happens under the hood when your employee asks the wrong question – and how stable are these railway guards against systematic harassment?
At the end we show how easy it is to strip the guardrails from an open-weights model. The uncomfortable part is not that this is possible – it's that attackers don't need our permission. They can do this every day, uncontrolled, unmonitored, without an ethics board. Where Anthropic's Mythos is tightly controlled, other nearly as powerful models can execute real attacks against our environments. Let's watch together what that actually looks like.
