Details
Running since June 2013, every month we have 1-2 talks focused on Security or Technology - followed by socialising with like minded people
- You can find out exactly what to expect about the night and the venue on the What We're About section of our Meetup page. Past talk details can be found the "Past Events" section.
- All people with any sort of an interest or level of knowledge in Security, Hacking and Emerging Technology are more than welcome to attend and feel free to bring like minded colleagues and friends.
CorkSec is made possible through generous sponsorship from our Platinum Sponsor TrendAI, Gold Sponsor CyberSkills , as well as our Silver and Bronze Sponsors featured prominently on the night.
Our talks come from our community so if you have an idea for a topic (anything for 10-60 minutes) please email us at **DefconCork@gmail.com** . Whether you are an experienced presenter, or presenting for your first time - CorkSec is a great venue for it - and we are happy to help you prepare and mentor you.
Doors open at 19:00 with talks starting at 19:15. Talks below
TALK 1: Yippee Ki Yay: Cowboy Incident Response with Local LLMs by Mark O'Sullivan
Armed with a few basic tools, a confident local model and some very questionable conclusions, can Roy spot the compromise and still make it to the office Christmas party?
Local LLMs offer security teams control over where their data goes, support for offline investigation and an alternative to token costs. But local inference brings its own constraints, finite memory, context limits and hardware bottlenecks. And, like their frontier counterparts, local models can sound entirely convincing while inventing facts.
Through a worked PCAP investigation, we’ll try different approaches to collecting evidence, testing competing explanations and identifying a possible compromise. We’ll examine what the model gets right, what it misses and what it confidently makes up.
Some topics covered:
- Running models local, quantisation, context windows, memory requirements, inference speed and trade-offs compared with frontier models.
- A hands on PCAP investigation moving from raw network data and manual prompting to bounded agents and structured evidence. (A live demo using an LLM what could go wrong!)
- LLMs in incident response: where they help, where they mislead, and why a plausible narrative is not the same as a supported conclusion.
TALK 2: TBC