What term describes anomalous patterns that could indicate systematic adversarial testing?

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

Answer: (D)

Recognizing signals of adversarial probing is about spotting patterns in interactions that suggest someone is systematically trying to learn how the model behaves. The term adversarial inference best fits this, because it describes attempts to infer a model’s internal properties and decision boundaries from observed outputs. When you see anomalous sequences of queries, crafted prompts, or repeated probing across different angles, these patterns indicate an effort to map vulnerabilities for later exploitation, which is exactly what adversarial inference covers. Other options describe separate phenomena: model drift involves natural shifts in data over time that can degrade performance; data poisoning is about injecting bad data into training; prompt injection is a specific attack where prompts are manipulated to coerce the model. But those don’t capture the idea of an attacker systematically probing to learn the model, which is what the question points to.