Adversaries sometimes compute gradients. Other times, they rob you.
Build your adversary flywheel.
Adversaries sometimes compute gradients.
[Read More]
What enables malicious models?
It's not just about the malicious models, to create a practical attack path, you need more ducks in a row.
Practicalities of Malicious Models
[Read More]
Introducing the Offsec ML Playbook v0.1
Enabling Red Teams to quickly leverage TTPs on ML infrastructure
Offsec ML Playbook
[Read More]
Using KServe to deploy malicious models
Weaponizing MLops for red teams and bounty hunters
This post builds upon my prior research into what red teams can do with ML environments. Now we look at using other components of common ML pipelines in our attacks as either pre or post exploitation targets.
[Read More]
Model Confusion - Weaponizing ML models for red teams and bounty hunters
How I hacked a bunch of companies via machine learning attacks.
This post accompanies my DEFCON31 AI Village talk - “You sound confused, anyways… Thanks for the jewels”.
[Read More]