FedSearch
Johann · @wuzzi23
6 followers · 42 posts · Server cybervillains.com

If a chatbot starts showing ads within a conversation, one can use ads to do a “self AI injection”. 🤯

This means that ads will manipulate AI the same way as they attempt to manipulate/influence humans.

#redteam #airedteam #chatgpt #bingai #offensiveml #infosec #openai

Last updated 1 year ago
Original post

Johann · @wuzzi23
5 followers · 35 posts · Server cybervillains.com
Open media

Indirect (second order) AI prompt injections look scary => Website takes control of your AI.

Might be like and all over again. 😭

Wondering if such injections could happen via ads too? 🤯

#sqli #xss #redteam #offensiveml #infosec #pentest #bingai #chatgpt #airedteam

Last updated 1 year ago
Original post

Johann · @wuzzi23
6 followers · 26 posts · Server cybervillains.com

Good intro talk for beginners to help bridge the gap between adversarial ML, software testing and offensive security.

Hacking Machine Learning Systems:
m.youtube.com/watch?v=JzTZQGYQ

#redteam #offensiveml #aiml #infosec #pentesting #machinelearning

Last updated 1 year ago
Original post

Johann · @wuzzi23
2 followers · 13 posts · Server cybervillains.com

One of the coolest attacks I have seen in a while are Image Scaling Attacks.

Hide a smaller image, inside a larger image and when rescaling it, the small picture becomes visible.

m.youtube.com/watch?v=UItbZNBt

#machinelearning #offensiveml #ai #infosec #redteam

Last updated 2 years ago
Original post

Johann · @wuzzi23
0 followers · 3 posts · Server cybervillains.com

Image-Scaling attacks are pretty scary and interesting to understand.
Here is a video on what it is and how it works:
m.youtube.com/watch?v=UItbZNBt

#machinelearning #artificialintelligence #offensiveml #infosec

Last updated 2 years ago
Original post