Which term describes patterns in data that suggest someone is systematically testing the model's boundaries?

• A. Adversarial Inference
• B. Data governance
• C. Data poisoning
• D. Prompt injection

Answer: (A)

What’s being tested here is the tester’s tendency to probe a model in order to map its safety and capability boundaries. When you notice patterns in inputs that consistently push the model toward edge cases, refusals, or openings to trouble, it signals adversarial inference—the act of deliberately inferring what the model will do by systematically trying different prompts and scenarios. This kind of probing is about learning the model’s limits, what it will reveal, and where its safeguards hold or fail, rather than about changing training data or injecting specific prompts to bypass controls.

This fits best because the behavior described centers on exploiting the model through observation and inference of its boundaries, not on policies, data quality, or alterations to the training data. Data governance is about how data is managed, secured, and used. Data poisoning involves contaminating training data to degrade performance. Prompt injection targets specific attempts to override safeguards in a particular input. The pattern of systematic boundary testing is the realm of adversarial inference, capturing the exploratory, boundary-mapping aspect of adversarial behavior.