AI Ethics, Privacy, Security, and Responsible Use
What to know
Responsible AI rests on principles: fairness, transparency/explainability, accountability, human oversight, privacy, and security. Know the privacy fundamentals: data minimisation, purpose limitation, consent, anonymisation vs pseudonymisation (pseudonymised data can often be re-identified), and retention limits. Familiarity with regulation at a conceptual level helps - for example GDPR-style obligations around personal data and lawful basis, and emerging AI-specific rules requiring risk assessment and disclosure.
Security risks include prompt injection (malicious instructions hidden in retrieved content), data leakage through prompts or logs, model inversion, and over-broad access. Governance tools: risk assessments, model cards, audit logs, access controls, and incident response.
Traps
Questions test whether you treat anonymisation as a checkbox - it is not, if re-identification is feasible. Beware answers that delegate accountability to the AI system; humans and organisations remain accountable. Also watch for conflating transparency (disclosing that AI is used) with explainability (justifying a specific decision).
Worked example
A hiring tool screens CVs. Responsible deployment requires: bias testing across demographic groups, disclosure to candidates, human review of rejections, retention limits on CV data, and logging of model versions. Rule of thumb: high-impact decisions about people need human oversight and documented justification.