Útmutatóhoz kapcsolódó kvíz · Közepes Szint
PII Redaction in LLM Pipelines Quiz
Compare PII detection and transformations, reversible tokens, false negatives, OpenAI Privacy Filter’s limits, and why redaction is not anonymization.
kérdés 1 -ból 8
What technique do PII redaction pipelines typically use to catch personal data with a predictable format, such as phone numbers or email addresses?
Tekintse meg az összes kérdést8 kérdéseket
A válaszok és magyarázatok megjelennek a kvíz kitöltésekor.
- What technique do PII redaction pipelines typically use to catch personal data with a predictable format, such as phone numbers or email addresses?
- Regular expression pattern matching
- Elnevezett entitás felismerés
- Homomorphic encryption
- Federated learning
- Why do pipelines add a named entity recognition (NER) step in addition to regex matching?
- To encrypt the detected entities
- To detect personal names, addresses, and organizations that have no fixed format
- To speed up regex matching
- To generate the lookup table automatically
- In a table-based reversible-tokenization design, where should the mapping from placeholder tokens to original values be kept?
- Inside the LLM's training data
- Embedded directly in the token text
- In a separate, access-controlled lookup kept in a trusted environment
- It is not stored anywhere
- What risk does a reversible-token lookup table create?
- It cannot be used to link placeholders
- It makes every record anonymous
- It removes the need to manage keys
- It becomes a sensitive asset that needs access controls and protection
- Why does the guide say redaction is not the same as anonymization?
- Because redacted text can sometimes still be re-identified from surrounding context clues
- Because redaction always requires encryption keys
- Because anonymization is illegal in most jurisdictions
- Because redaction only works on numeric data
- Why can a stable pseudonym remain privacy-sensitive even when it replaces a direct identifier?
- The original identity in plaintext
- It can preserve links across records and may aid re-identification
- The exact person’s real phone number
- A guarantee that the replacement is safe to contact
- What should be checked after a redaction pass before content enters logs or a downstream model?
- Only whether the prompt is shorter
- Whether the language model can infer every placeholder
- Whether serialization, errors, and downstream traces still contain sensitive spans
- Nothing; one detector guarantees complete removal
- How does the guide characterize OpenAI Privacy Filter?
- As a general compliance certification
- As an API that automatically redacts all requests
- As a reversible token vault for all cloud data
- As an open-weight text PII detection and masking aid with known limits