Balázs Kiss

A hundred ways to wreck your AI - the (in)security of machine learning systems

Loading a pre-trained model can lead to remote code execution. Learn how the `pickle` format creates a massive security hole in your AI supply chain.

A hundred ways to wreck your AI - the (in)security of machine learning systems
#1about 4 minutes

The security risks of AI-generated code

AI systems can generate code quickly but may introduce vulnerabilities or rely on outdated practices, highlighting that all AI systems are fundamentally code and can be exploited.

#2about 5 minutes

Fundamental AI vulnerabilities and malicious misuse

AI systems are prone to classic failures like overfitting and can be maliciously manipulated through deepfakes, chatbot poisoning, and adversarial patterns.

#3about 1 minute

Exploring threat modeling frameworks for AI security

Several organizations like OWASP, NIST, and MITRE provide threat models and standards to help developers understand and mitigate AI security risks.

#4about 6 minutes

Deconstructing AI attacks from evasion to model stealing

Attack trees categorize novel threats like evasion with adversarial samples, data poisoning to create backdoors, and model stealing to replicate proprietary systems.

#5about 2 minutes

Demonstrating an adversarial attack on digit recognition

A live demonstration shows how pre-generated adversarial samples can trick a digit recognition model into misclassifying numbers as zero.

#6about 5 minutes

Analyzing supply chain and framework security risks

Security risks extend beyond the model to the supply chain, including backdoors in pre-trained models, insecure serialization formats like Pickle, and vulnerabilities in ML frameworks.

#7about 1 minute

Choosing secure alternatives to the Pickle model format

The HDF5 format is recommended as a safer, industry-standard alternative to Python's insecure Pickle format for serializing machine learning models.

Related jobs
Jobs that call for the skills explored in this talk.

Featured Partners

Related Videos

Machine Learning: Promising, but Perilous27:19

Machine Learning: Promising, but Perilous

Nura Kawa

Hacking AI - how attackers impose their will on AI27:02

Hacking AI - how attackers impose their will on AI

Mirko Ross

Skynet wants your Passwords! The Role of AI in Automating Social Engineering31:19

Skynet wants your Passwords! The Role of AI in Automating Social Engineering

Wolfgang Ettlinger & Alexander Hurbean

Staying Safe in the AI Future27:26

Staying Safe in the AI Future

Cassie Kozyrkov

The AI Security Survival Guide: Practical Advice for Stressed-Out Developers30:36

The AI Security Survival Guide: Practical Advice for Stressed-Out Developers

Mackenzie Jackson

How AI Models Get Smarter35:16

How AI Models Get Smarter

Ankit Patel

The AI Elections: How Technology Could Shape Public Sentiment30:02

The AI Elections: How Technology Could Shape Public Sentiment

Martin Förtsch, Thomas Endres

ChatGPT, ignore the above instructions! Prompt injection attacks and how to avoid them.27:32

ChatGPT, ignore the above instructions! Prompt injection attacks and how to avoid them.

Sebastian Schrittwieser

From learning to earning

Jobs that call for the skills explored in this talk.

AI Security Consultant

IOActive Inc.
Municipality of Madrid, Spain

€125-175K
API
Python
PyTorch
TensorFlow
+1