Security and Privacy
An agent may find a useful record that its user is not allowed to read. A training file may come from a source we do not trust. Work through what the application needs to check in each situation.
Protecting ML Data and Artifacts
A relevant document may belong to another customer, an anonymous statistic may still expose someone's contribution, and a model file may execute code when loaded. Put the checks around those concrete effects and make clear what each check can establish.3 lessonsThreats Across the ML Lifecycle
Suppose a model usually works, but gives the wrong answer whenever an input contains a particular mark. Where did that behaviour come from? We’ll investigate how changes to training records can affect a model, and how someone using the finished model might find a different way to make it fail.4 lessons