- Last Week As A vCISO
- Posts
- BSides Day 2 Summary - AI
BSides Day 2 Summary - AI
Fun facts and AI Talks
BSidesSF is an amazing event. It comes and goes so quickly, but it just has a special place in my heart. It’s the place where I first played a CTF and made some lifelong friends in the industry.
Fast forward to today it’s like a mini, yet grown up, Def Con with villages, talks, and more. A worthwhile event.
I found it folks! The single pane of glass to solve all our problems!
Table of Contents
Some facts about BSidesSF
$4435 was raised for charities via T-Shirt Sales
BSidesSF is the most expensive Besides in the world as it takes place in one of the most expensive cities in the world
10+ lbs of jelly beans was consumed by staff
Headshots at BSidesSF
Something new at BSidesSF is headshots! Yes, they had free headshots all day! A great opportunity to stand in an awkward position and show the world your best side.
AI Talks
Considering the theme of the conference was “You can’t spell dystopia without AI”, I found it was befitting to see some talks on AI and Security.
Talk: Insane In The Supply Chain: Threat Modeling For Attacks on AI Systems
By Marta Janus, Eoin Wickens
Link: https://sched.co/1abFH
My main takeaway from this talk is that AI Security is like taking software security back to 2000.
Back to basics like scanning for signatures, checksums, and other table stakes these days.
AI Security is like taking software security back to 2000.
Back to basics like scanning for signatures, checksums, and other table stakes these days.
— Ayman Elsawah ☕👋🏼 (@coffeewithayman)
9:46 PM • May 5, 2024
Takeaways
Security Of ML Formats Are Insecure By Design
Insecure Code
Often vulnerable by design
No digital signatures / certs
No integrity checks
No malware scanning
Not quite considered “software”
Model Hijacking Issues
People want a plug and play solution for their models
Check signatures from the author
Scan for malware or erroneous data
Deploy in a sandbox first
Monitor Constantly
Don’t allow direct access to models
I highly recommend watching this talk when published if you are building anything with AI models!!
Talk - AI: Best Janitor or Worst Superhero?
By Adrian Sanabria
Link: https://sched.co/1abFl
The title of this talk reminded me of the Space Quest series from Sierra games, where a hapless janitor named Roger Wilco accidentally saves the galaxy.
Adrian always has a knack for framing things. I love his perspective and it always adds great color to some of our problems in the industry.
Takeaways and Quotes:
People sometimes put pressure on the AI for it perform better
“I will lose my job if you don’t do…”
Stroking its ego! “You are amazing…”
Why make AI do high stakes items, when it can just do simple and low stake things we need in security
Pumping pcaps and analyzing event data
Fill out security questionnaires
People are copying and pasting anyway!!
Board deck slides?
AI Made some suggestions and created a slide!!
Help desk tickets
Password resets
This slide was ENTIRELY created by AI
Villages
Below is the full list of villages at the con:
Adversary Village
Bug Bounty Village
Career Village
Crypto & Privacy Village
Hardware Challenge Village
Personal Security Village
API Security Village
Capture the Flag
Cloud Village
Embedded Systems Village
Lockpick Village
AI Village (Of course!)
From the Embedded Systems Village
Reply