Which statement best describes differential privacy in SecAI+?

• A. It guarantees zero error on training data.
• B. It bounds influence of any single record on outputs to protect privacy.
• C. It eliminates the need for data governance.
• D. It increases model capacity.

Answer: (B)

Differential privacy aims to limit how much any single individual's data can sway the results of a computation. In SecAI+, this is achieved by mechanisms that add calibrated randomness or otherwise constrain how much a single record can affect outputs. The idea is that whether or not one person’s data is included slightly changes the result, but the change is bounded and statistically insignificant enough to protect privacy. This bound is formalized with privacy parameters that control how much influence a single record can have, ensuring the presence or absence of one dataset entry doesn’t reveal sensitive information. That’s why this statement best describes differential privacy: it bounds the influence of any single record on outputs to protect privacy. The other options don’t fit because zero error on training data isn’t guaranteed with differential privacy due to the added noise, data governance isn’t eliminated by this technique, and increasing model capacity isn’t a stated outcome of applying differential privacy.