AI Excessive Agency
Investigating how AI systems can exceed their intended boundaries and how insecure agent design leads to vulnerabilities — explored through a real-world example.
Read →cat ./blog_posts.log
Stories
Notes on offensive security, AI red teaming, and the occasional idleness.
Evasion Attack on AI Classifier
How adversarial examples can be crafted to evade AI classification systems. We walk through the theory behind evasion attacks and set up an experimental environment using the Adversarial Robustness Toolbox (ART).
Read →