During a review of the logs, a cybersecurity analyst notices that the data inputs recorded for a machine learning model used in fraud detection exhibit an unusual pattern that significantly deviates from the expected data format and ranges. The anomaly led to a drop in detection accuracy, and subsequent investigations pointed to external manipulation. What type of vulnerability is most likely being exploited in this scenario?
Data poisoning is a technique where an attacker introduces corrupt or malicious data into a system's data set to manipulate the behavior of a machine learning model, reduce its effectiveness, or cause it to make incorrect predictions. While parameter tampering involves manipulating inputs, data poisoning is the specific term for an attack that corrupts the training or input data for an ML model to degrade its performance. Buffer overflow is a memory corruption vulnerability, and cross-site request forgery is an attack that tricks a user into submitting a malicious request; neither fits the scenario.
Ask Bash
Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.
What is data poisoning in machine learning?
Open an interactive chat with Bash
How can an organization detect and prevent data poisoning attacks?
Open an interactive chat with Bash
How is data poisoning different from adversarial attacks on inputs during inference?