UPSC Prelims Practice Questions — How AI training models could compromise privacy

Q1. With reference to the Digital Personal Data Protection Act, 2023, consider the following statements: Which of the statements given above are correct?

  1. Personal data may be processed only for a lawful purpose, either with the Data Principal's consent or for certain legitimate uses specified in the Act.
  2. Obligations such as Data Protection Impact Assessments and independent audits apply to every Data Fiduciary irrespective of the volume or sensitivity of the data it processes.
  3. A Data Fiduciary becomes a Significant Data Fiduciary through Central Government notification, having regard to factors such as the volume and sensitivity of data and the risk to Data Principals.
  4. The Data Protection Board of India derives its existence from the Act and is established by the Central Government, rather than being a constitutional body.
  • A. 3 and 4 only
  • B. 2, 3 and 4 only
  • C. 1 and 2 only
  • D. 1, 3 and 4 only

Q2. Consider the following findings: Studies show that when an AI model is queried with 'anonymised' data, it can return personal details it memorised during training, and these details can be used to identify individuals and build profiles of them. Which of the following inferences can be drawn from the above?

  1. A model's outputs can disclose information about individuals that was not contained in the queries put to it.
  2. Privacy risk arising from an AI model can persist after the model's training has been completed.
  3. Every AI model trained on personal data will necessarily reveal that data whenever it is queried.
  • A. 1 and 2 only
  • B. 1 only
  • C. 1, 2 and 3
  • D. 2 and 3 only

Q3. Consider the following statements: Which of the following is correct in respect of the above statements?

  1. Restricting access to an AI model's training dataset does not by itself eliminate the risk that personal data contained in that dataset will leak.
  2. Personal data memorised during training can become embedded in a model's weights and be reproduced in its outputs.
  • A. Both Statement-I and Statement-II are correct and Statement-II explains Statement-I
  • B. Both Statement-I and Statement-II are correct, but Statement-II does not explain Statement-I
  • C. Statement-I is correct, but Statement-II is incorrect
  • D. Statement-I is incorrect, but Statement-II is correct