Understanding the Threats and Attacks on Data Science Applications and Models
summary
Shows concrete threats: data poisoning via fake tweets and millisecond buy‑cancel orders that trick trading models, sensor spoofing in manufacturing, and streaming sources being easier to corrupt than batch. Lists vulnerable infrastructure: Kafka Spark Flink Airflow Kubernetes, warns spark is off by default for many deployments, and calls out Jupyter exposing remote code execution plus pickle deserialization as a common exploit. Notes S3 misconfigurations, credential leakage, and type/encoding mismatches that break deduplication and input filters.
AI-written summary · about
chapters (15)
- 0:00Understanding the Threats and Attacks on Data Science Applications
- 0:25Biggest Blindspot in Enterprise Security
- 2:57The Attack Surface of a Data Science Application
- 4:16Attacking an Algorithm's Data
- 11:45Securing Data Science Infrastructure
- 15:04Attacking Data Science Engineers
- 17:06Attacking Data Science Developer Tools
- 19:51Attacking Data Cleansing and Preparation Logic
- 22:34Processing Components
- 25:09Input Problems Unique to Streaming Apps
- 25:35Ensuring that Your Data is Bounded
- 27:29Filter Bypass Issues
- 28:31Mismatched Character Sets
- 30:22Knowing All of Your Potential Data Uses
- 32:23Questions and Contact
related talks
46mInsane in the Supply Chain: Threat modeling for...
45mThreat Modeling in the Age of AI
43mHacking and Securing ML Systems and Environments
37mBeyond the Basics: Hacking and Securing Machine Learning Systems and Environments
12mSecuring AI-Driven Finance
02h 13mUnderstanding Security Threats Against Machine/Deep Learning Applications
46mAvoiding Common Design and Security Mistakes in Cloud AI/ML Environment
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production