Cuestionario vinculado a una guía · duro Nivel

Cuestionario de privacidad e inteligencia artificial

Prueba los flujos de datos, la privacidad diferencial, la inferencia de membresía, la reidentificación indirecta, la eliminación y los controles de riesgo de privacidad.

Rutas de guía relacionadasPrivacidad de la IA
Pregunta 1 de 8

Which statement best distinguishes security from privacy?

Vista previa de cada pregunta8 preguntas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

  1. Which statement best distinguishes security from privacy?
    • Security and privacy are identical concepts
    • Encryption alone establishes consent
    • Secure storage can still be a privacy problem if data are excessive or used unexpectedly
    • Privacy only concerns passwords
  2. In a differential privacy analysis, what does epsilon influence?
    • The number of records in the input
    • The privacy-accuracy tradeoff, with smaller epsilon generally giving stronger privacy
    • The model’s training temperature
    • The encryption key length
  3. A membership-inference attacker asks whether a person’s record was in training. What is being tested?
    • Whether a user can reset a password
    • Whether a record is duplicated in a database
    • Whether a prompt contains a name
    • Whether the model output leaks information about training-set membership
  4. Why can removing names fail to make a dataset anonymous?
    • Other attributes or combinations can re-identify people
    • A name is the only possible identifier
    • Data cannot be linked across tables
    • Removing names always satisfies every privacy goal
  5. Which data practice best follows the guide’s privacy principle?
    • Keep every prompt indefinitely for any future use
    • Collect only information needed for the stated task and set a retention period
    • Copy all source records into each model call
    • Treat local processing as proof of no privacy risk
  6. Why should a team inspect embeddings and indexes after deleting an uploaded document?
    • Deleting a document automatically removes every copy everywhere
    • Embeddings contain no data relationships
    • Derived artifacts may still reveal information from the original data
    • An index is only a user interface
  7. What should a team document about AI data processing?
    • Only the model’s public name
    • The number of support tickets
    • A generic privacy slogan
    • Which providers and internal services handle each data category
  8. Why does an assistant’s unsupported claim that data were deleted not suffice?
    • The team should test deletion and account isolation through the actual application
    • A chatbot response is a deletion receipt
    • Data deletion is never relevant
    • The assistant can inspect all vendor systems