SECAI Core · phase 15 of 15
Model skewing
When the behavior of a machine learning model is altered through biased or manipulated training or feedback data
The Explain card
- Plain English
- Model skewing is when a model's behaviour is altered by biased or manipulated training or feedback data, pushing its decisions in a direction the attacker wants.
- Example
- A spam filter retrains on user feedback. An attacker creates many accounts that mark their spam as "not spam", and over weeks the filter learns to let their campaigns through.
- Why it matters
- Any model that learns from production data or user feedback can be steered by whoever controls that data. Defenders must validate feedback, limit its influence and watch for behavioural shifts.
- Hook
- Feed the model lies long enough and it believes them.
Where it sits in the deck
Phase 15: MLOps, Continuous Delivery, and Operational Resilience
Close the loop — operational practices, drift management, and delivery discipline that keep deployed AI systems healthy, current, and continuously improving.
