A bank is configuring a customer-facing generative AI assistant and wants controls that express the privacy and security principle, as opposed to controls that belong to other responsible AI principles. Which two measures most directly uphold the privacy and security principle? Select TWO.
- AMinimise the personal data the assistant retains and purge transcripts once a query is resolved. Correct
- BHarden the assistant against prompt-injection attempts that try to extract its hidden instructions or data. Correct
- CProvide a clear route for a customer to reach a human agent when the assistant cannot help.
- DDisclose to customers that they are interacting with an AI assistant rather than a person.
- EMonitor the assistant for output drift and retrain it when its answer quality degrades over time.
Why A is correct: Limiting and purging retained personal data directly serves the privacy half of the privacy and security principle.
Why B is correct: Defending against adversarial inputs that exfiltrate data is a core expression of the security half of the principle.
Why C is wrong: A human-escalation route is good design but it expresses human-centricity, not the privacy and security principle.
Why D is wrong: Disclosing the AI nature of the tool is tempting as a safeguard but it serves transparency, not privacy and security.
Why E is wrong: Drift monitoring is sound practice but it serves the safety and reliability principle rather than privacy and security.