AI assistants can be useful for cybersecurity, but using them often means sending source code, logs, vulnerability details, and other sensitive data to infrastructure operated by a small number of mostly US-based providers.

This talk explores using open-weight LLMs locally for cybersecurity tasks on accessible hardware. Local inference provides data and infrastructure sovereignty: sensitive material stays within an environment you control, without depending on a foreign cloud provider, its policies, or availability.

We will look at practical security applications, hardware requirements, quantization, and the additional freedom open-weight and less-restricted models provide for inherently dual-use security research.

And what better place to demonstrate independence from the cloud than on a ship? The demos will run locally on hardware brought aboard, with the network disabled. No cloud APIs or connection back to shore required.

---

Unto Karjalainen

Unto Karjalainen is a DevSecOps Architect working where software engineering, cloud, cybersecurity, and regulation meet, with a growing focus on practical and secure AI adoption.

---