Table of Contents
- 1. Lesson Overview
- 2. Learning Outcomes
- 3. Before You Begin
- 4. Retrieval Practice
- 5. Opening Scenario
- 6. Privacy, Confidentiality, Security, and Data Protection
- 7. Information Classification
- 8. Authority, Ownership, and Stewardship
- 9. Purpose Limitation
- 10. When Not to Supply Information
- 11. PROTECT Walkthrough
- 12. Process Walkthrough
- 13. Worked Example
- 14. Guided Learning Activities
- 15. Exercises
- 16. Homework Problems
- 17. Mini-Project
- 18. Self-Check Quiz
- 19. Assignment Connections
- 20. Key Takeaways
- 21. Key Terms
- 22. Looking Ahead
- 23. Sources & Further Reading
1. Lesson Overview
This lesson establishes the foundational information-governance concepts you will need throughout Unit 8 and your professional life. You will learn to distinguish privacy, confidentiality, security, and data protection—four related but distinct ideas that are often confused. You will then develop the ability to classify information provisionally using a set of nine categories, and to identify who owns, governs, or stewards information. Most importantly, you will learn to apply purpose limitation: the principle that information collected for one purpose may not be used for an incompatible purpose, even if you have access to it.
This lesson prepares you to ask—and answer—the critical question: “Does my access to this information create authority to use it in this AI system for this purpose?” The answer is often no, and knowing when to stop is a core responsible-AI skill.
2. Learning Outcomes
By the end of this lesson, you should be able to:
- Define privacy, confidentiality, security, data protection, consent, authority, purpose limitation, and incident.
- Distinguish privacy from confidentiality, security from privacy, and data protection from each.
- Classify information provisionally as public, internal, personal, sensitive personal, confidential, proprietary, privileged, security-sensitive, or culturally governed.
- Explain why final classification depends on applicable policy, law, contract, professional duty, or community authority.
- Identify information owners, stewards, custodians, users, reviewers, and decision authorities.
- Explain why possessing or accessing information does not create permission to use it in an AI system.
- Determine whether an AI-supported use is compatible with the authorized purpose.
- Apply purpose limitation to a synthetic case and recommend whether information may be supplied, generalized, replaced, or excluded.
3. Before You Begin
- Assigned reading: Chapter 10: Privacy, Security, and Data Protection (sections on privacy, confidentiality, security, information classification, authority, consent, and purpose limitation).
- Prerequisite knowledge: You should be able to explain the AI Paradox, apply ORBIT, distinguish AI models/applications/systems, explain context windows and retrieval, apply PCAFE-V, SCOPE, PARTNER, distinguish human authority from AI capability, identify bias and fairness concerns, apply FAIRER terminology, and identify contestability, remedy, escalation, and stop conditions (Units 1–7).
- Optional approved AI tool: If you have access to an institutionally approved AI system, you may use it with synthetic, public, or generalized information only. Do not enter real or sensitive information.
- Supplied-output alternative: If you do not use a live system, supplied prompts, synthetic records, and system descriptions are provided throughout. You will not be disadvantaged.
- Academic integrity: All work must be your own. When you use AI tools, document your use according to your institution’s policy.
- Accessibility: All activities are available in text; diagrams are described; all answers are in expandable details elements.
- Estimated preparation time: 20–30 minutes.
4. Retrieval Practice: Concepts from Units 3–7
These questions draw on Units 3–7. Try to answer before expanding the feedback.
Question 1: What is the AI Paradox, and how does it relate to responsible use?
Show answer
Answer: The AI Paradox is the tension between AI’s remarkable capabilities and its significant risks, limitations, and uncertainties. It reminds us that powerful systems require more human oversight, not less, because their opaqueness, fallibility, and potential for harm scale with their capability.
Explanation: In information governance, the paradox means that because AI can process vast amounts of information and generate plausible outputs, we must be more careful about what we supply, who can access it, and what the system does with it—not less.
Question 2: From Unit 6, what is the primary difference between human authority and AI capability?
Show answer
Answer: Human authority is the legitimate power to make decisions, take actions, and accept accountability based on values, ethics, law, and professional judgment. AI capability is the functional ability of a system to perform a task, generate content, or produce an output. A system may be capable of generating a sensitive document, but only a human with authority may decide whether to generate, use, or share it.
Explanation: This distinction is foundational for information governance: a system may have access to information (technical capability), but that does not give it—or the user—authority to use it. Authority resides with people and institutions, not with the AI.
Question 3: What is the purpose of the FAIRER framework introduced in Units 6–7, and how might it relate to privacy and security?
Show answer
Answer: FAIRER is a framework for evaluating AI systems with attention to Fairness, Accountability, Inclusivity, Responsibility, Explainability, and Robustness. In privacy and security, FAIRER reminds us that information protection is not neutral: it affects people differently. For example, privacy risks may be higher for marginalized groups, and security measures must not create accessibility barriers.
Explanation: When you classify information and apply purpose limitation, you must consider who is affected, whether they have a say, and whether the protection measures themselves are fair and inclusive. FAIRER helps keep people—not just data—at the centre.
Question 4: In Unit 5, why is it important to distinguish context from instruction in a prompt?
Show answer
Answer: Context is the information the system uses to understand the task; instruction is the directive about what to do. Distinguishing them helps prevent the system from treating source content as authoritative commands. This distinction is directly relevant to prompt injection (Lesson 8.3) and to ensuring that unauthorized instructions in retrieved material do not override the user’s intent.
Explanation: In information-governance terms, context may contain sensitive information; instruction tells the system how to process it. If you cannot separate them, you may inadvertently expose information or follow untrusted directions.
5. Opening Scenario: "The Internal Report"
Context: You work as a policy analyst for a mid-sized city. You have access to an internal database that contains a confidential draft report on a proposed new housing policy. The report includes detailed demographic data, budget projections, and feedback from community consultations. The report was prepared for the city council’s internal review and is marked “DRAFT – INTERNAL USE ONLY – NOT FOR DISTRIBUTION.”
You are experimenting with a free, public AI system to help you summarise the key arguments for a presentation you are preparing. You copy several paragraphs from the report into the AI prompt, including specific budget numbers and a quote from a community member. You do not include the person’s name, but you include their role (“a local business owner”) and the neighbourhood.
Questions to consider:
- Does your ability to access the report create authority to use it in this AI system?
- What categories of information might be present in the report?
- What was the original purpose of the report? Is the proposed AI use compatible with that purpose?
- What alternatives exist if you need to summarise the arguments without exposing the report?
Show guided analysis
1. Access vs. authority: No. Access is a technical permission (you can read the file). Authority is a legal, policy, or ethical permission to use it for a specific purpose. Your employment may give you access for internal policy review, but it does not automatically give you authority to copy the content into an unapproved public AI system. The system’s terms of service, the city’s information-management policy, and the report’s classification all matter.
2. Information categories: The report likely contains confidential (marked internal use), internal (city business), personal (quotes from community members, even if names are removed), sensitive personal (if demographic data relates to vulnerable populations), proprietary (budget models may be city assets), and possibly security-sensitive (if it includes infrastructure or emergency planning).
3. Original vs. proposed purpose: Original: internal review and council decision-making. Proposed: summarising for a presentation (likely still internal, but the means—using a public AI system—changes the context). The purpose may be compatible if the summary stays internal and does not reveal the report, but the method (public AI) likely violates the report’s classification and the city’s policies.
4. Safer alternatives: Use a city-approved AI system if one exists; work with a synthetic or aggregated version of the data; use a non-AI summarisation method; consult with a privacy or records officer; or request a redacted version of the report for your specific purpose.
6. Privacy, Confidentiality, Security, and Data Protection
Privacy
Privacy involves appropriate control, governance, use, access, disclosure, and protection of information about people. It is not simply “keeping secrets” or “removing names.” Privacy is about human dignity, autonomy, and trust: people have a legitimate interest in how information about them is collected, used, shared, and protected.
Confidentiality
Confidentiality is an obligation or expectation that information will be accessed, used, and disclosed only as authorised. Sources of confidentiality may include:
- Law (e.g., privacy legislation)
- Contract (e.g., non-disclosure agreements)
- Professional duty (e.g., lawyer–client, doctor–patient)
- Institutional policy (e.g., university records)
- Research ethics (e.g., participant data)
- Employment responsibility (e.g., workplace confidentiality)
- Community protocol (e.g., Indigenous knowledge governance)
- Explicit agreement (e.g., a signed promise)
Security
Security protects information and systems against:
- Unauthorised access
- Unauthorised disclosure
- Alteration
- Loss
- Destruction
- Misuse
- Unavailability
- Malicious or accidental action
Security is often described using the CIA triad (conceptual):
- Confidentiality: ensuring information is accessible only to authorised parties.
- Integrity: ensuring information is accurate and unaltered.
- Availability: ensuring information and systems are accessible when needed.
Data Protection
Data protection is the broad umbrella of policies, legal requirements, governance processes, and technical or organisational controls used to manage information responsibly throughout its lifecycle. It includes privacy, confidentiality, security, and more—such as purpose limitation, data minimisation, retention, deletion, and accountability.
| Concept | Core focus | Primary concern | Example |
|---|---|---|---|
| Privacy | Control over personal information | Autonomy, dignity, trust | You decide who sees your medical record. |
| Confidentiality | Obligation to protect information | Duty, trust, authorised use | A lawyer must not disclose client information. |
| Security | Protection against threats | Confidentiality, integrity, availability | Encrypting data to prevent interception. |
| Data protection | Full lifecycle governance | Policy, law, process, and accountability | An organisation’s privacy policy, retention schedule, and breach-response plan. |
Consider the scenario from Section 5. Is the issue primarily privacy, confidentiality, security, or data protection?
Show analysis
It is all four, but the primary concern is confidentiality (the report is marked internal and confidential) and data protection (the city’s policies and processes are being violated). Privacy is involved because the report contains community members’ feedback. Security is involved because a public AI system is not an approved secure channel. The problem is not just one category—they overlap.
- Privacy, confidentiality, security, and data protection are distinct but overlapping concepts.
- Privacy is about people’s interests; confidentiality is an obligation; security is a means; data protection is the whole system.
- Understanding the distinctions helps you determine which controls apply to a given situation.
7. Information Classification
Information classification is a provisional exercise. The final classification depends on applicable policy, law, contract, professional duty, or community authority. In this course, we use nine provisional categories. They may overlap.
Public
Authorised for unrestricted public access.
Internal
Intended for use within a defined organisation or group.
Personal
Relating to an identifiable person.
Sensitive personal
Personal information that may create greater risk if misused, exposed, or inferred.
Confidential
Subject to restrictions on access, use, or disclosure.
Proprietary
Owned or controlled as a business, organisational, technical, or intellectual asset.
Privileged
Protected by a recognised legal or professional privilege (where applicable).
Security-sensitive
Could enable unauthorised access, exploitation, circumvention, or harm.
Culturally governed
Information, knowledge, language, records, or practices governed by community authority, protocols, or collective rights.
Look at the opening scenario (Section 5). Which categories apply to the draft report?
Show sample classification
- Confidential – marked internal use only.
- Internal – city business.
- Personal – community member quotes.
- Sensitive personal – if demographics identify vulnerable groups.
- Proprietary – budget models are city assets.
- Security-sensitive – if it includes infrastructure or emergency details.
- Not public, not privileged (unless legal advice), and not culturally governed (unless it includes Indigenous knowledge, which it does not in this case).
Note: This is a provisional classification. The city’s records office would make the final determination.
- Nine provisional categories: public, internal, personal, sensitive personal, confidential, proprietary, privileged, security-sensitive, and culturally governed.
- Categories may overlap.
- Classification is provisional; final determination depends on the governing authority.
9. Purpose Limitation
Purpose limitation requires that information be used only for authorised and specified purposes. It is a cornerstone of privacy and data protection law, policy, and ethics. You cannot collect information for one purpose and then use it for another, incompatible purpose—even if you could technically do so.
Guiding questions for purpose limitation
- Why was the information collected? What was the original, documented purpose?
- Who authorised its use? Who gave permission, and what were the limits?
- Is AI processing compatible with that purpose? Does the proposed AI use serve the same or a directly related purpose?
- Is a new authorisation required? If the purpose is different, you may need fresh consent or a new policy basis.
- Will the information be used for training, improvement, analytics, or another secondary purpose? If so, that is a new purpose and likely requires new authority.
- Can the task be completed with less information? If you can achieve the purpose with less information, you should—that is data minimisation (covered in Lesson 8.2).
In the opening scenario, what is the original purpose of the report? What is the proposed AI purpose? Are they compatible?
Show analysis
Original purpose: Internal review and decision-making by the city council.
Proposed AI purpose: Summarising key arguments for a presentation.
Compatibility: The substantive purpose (informing decision-making) may be compatible, but the method (using a public AI system) introduces new risks and a possible secondary use (e.g., the AI provider may use the input for model improvement, depending on the system’s terms). Even if the provider does not use it for training, the act of copying into a system that is not approved for confidential information may violate the city’s policy. Therefore, the proposed use is not authorised without additional safeguards and approval.
- Purpose limitation restricts use to the authorised purpose.
- A new purpose requires new authorisation.
- Secondary use (e.g., training) is a new purpose.
- Using an AI system may introduce a secondary purpose even if the substantive goal seems compatible.
10. When Not to Supply Information
You must stop and not supply information to an AI system when:
- You lack clear authority to use it for the proposed purpose.
- The information is classified as confidential, personal, sensitive, proprietary, privileged, security-sensitive, or culturally governed, and you have not obtained the required authorisation.
- The AI system is not approved for that category of information (e.g., a public system for confidential data).
- You are unsure about the system’s memory, retention, or secondary use.
- The purpose limitation test fails.
- Your organisation’s policy explicitly prohibits the use.
- You cannot verify where the information will be stored or who can access it.
- Multiple conditions require you to stop and not supply information.
- When in doubt, stop and seek guidance.
- The safest choice is often to avoid the AI system entirely.
11. PROTECT Walkthrough
Definition: Every use of information must be grounded in a legitimate, authorised purpose and supported by appropriate authority (legal, policy, consent, or other basis).
- Guiding questions: What is the purpose? Who authorised it? Is the proposed AI use compatible with that purpose? What limits apply?
- Example: A student uses a course-approved AI system to generate study notes from publicly available course materials. The purpose is learning, the authority is the course policy, and the system is approved. This is permissible.
- Common mistake: Assuming that because you can access information (e.g., through a shared drive), you have authority to use it in any way.
- Control: Document the purpose and authority before using any information in an AI system.
- Escalation/stop condition: If purpose or authority is unclear, stop and seek clarification from the responsible authority.
- Residual limitation: Even with authority, you must still apply other PROTECT elements (classification, minimisation, etc.).
Definition: Information must be reviewed and classified to determine its sensitivity, risks, and appropriate handling.
- Guiding questions: What information is involved? What categories apply (public, internal, personal, confidential, etc.)? Who determines the classification? Do categories overlap?
- Example: Before using a client list in an AI system, you review it and classify it as confidential and personal. You then determine that it cannot be entered into a public system.
- Common mistake: Treating all information as “public” or “low risk” by default.
- Control: Use a classification checklist or framework to review all information before use.
- Escalation/stop condition: If you cannot classify the information with confidence, or if it appears to fall into a restricted category, stop and seek guidance.
- Residual limitation: Classification is provisional; final classification depends on the governing authority.
Definition: Information is not owned by the person who uses it; it is owned, governed, or stewarded by an authoritative entity that defines how it may be used.
- Guiding questions: Who owns or governs this information? Who is the steward or custodian? Who has decision authority? Who can authorise a new use?
- Example: A research dataset may be owned by a university, stewarded by a principal investigator, and subject to ethics committee authority. You, as a research assistant, may access it but cannot use it for a new project without approval.
- Common mistake: Acting as if you own information because you created it or have access to it.
- Control: Map the authority structure before using information.
- Escalation/stop condition: If ownership or stewardship is unclear, stop and contact the responsible office.
- Residual limitation: Even with ownership identified, you must still obtain authorisation for the specific use.
12. Process Walkthrough: Information Collection to Authorised Use
This walkthrough shows the decision process you should follow before supplying any information to an AI system. It integrates the core concepts from this lesson.
- Identify the information. What is it? Where did it come from?
- Determine the purpose. Why was it collected? What is its authorised use?
- Classify the information provisionally. Use the nine categories. Note overlaps.
- Identify the governing authority. Who owns it? Who stewards it? Who decides?
- Define your proposed AI use. What are you trying to do? What system will you use?
- Test purpose limitation. Is the proposed use compatible with the original purpose? If not, stop or seek new authorisation.
- Check authority. Do you have the necessary authority to use the information this way? If not, stop or seek approval.
- Consider alternatives. Can you use less information? Synthetic data? A different system?
- Make a preliminary decision: Proceed, proceed with safeguards, redesign, use a non-AI alternative, seek specialist review, pause, or stop.
- Document your reasoning. Record what you decided and why—this is evidence for Assignment 3.
Apply the process to the opening scenario. Which step would stop you?
Show analysis
The process would likely stop at step 6 (purpose limitation) and step 7 (authority). The proposed use (copying confidential report text into a public AI system) is not compatible with the original purpose (internal review), and you do not have authority to use the information in that way. The correct decision is to stop and seek a safer alternative.
13. Worked Example: Student Feedback Analysis
Context: You are a teaching assistant for a large undergraduate course. The course coordinator has asked you to analyse anonymous student feedback from course evaluations to identify common themes for next year’s course improvement. The feedback is stored in a university-approved system and is marked “Internal – Research and Teaching Improvement.” You have access to the feedback as part of your role.
Proposed AI use: You want to use a public AI system (e.g., a free chatbot) to summarise the feedback and identify recurring themes. You plan to copy the text of the comments into the system, but you will remove student names.
Worked Analysis
- Information: Student evaluation comments (text).
- Original purpose: Course improvement and teaching quality review (internal).
- Provisional classification: Internal (university use), Personal (identifiable? The feedback is anonymous, but may still contain indirect identifiers), Confidential (not for public release).
- Governing authority: The university (through the course coordinator and the institutional research office).
- Proposed AI purpose: Summarising feedback for course improvement.
- Purpose limitation test: The substantive purpose is compatible (teaching improvement). However, the method (public AI system) introduces a secondary use because the AI provider may retain or use the input. The university’s policy may prohibit using student feedback in public systems. Result: Incompatible without additional safeguards.
- Authority: You have access to the feedback, but you do not have authority to copy it into a public AI system. You would need approval from the course coordinator and the privacy office. Result: Not authorised.
- Alternatives: (a) Use a university-approved AI system if one exists; (b) use a non-AI summarisation method; (c) create a synthetic or aggregated version of the feedback; (d) ask the coordinator to provide a redacted version for this specific purpose.
- Preliminary decision: Do not proceed with the public AI system. Seek a safer alternative or obtain explicit authorisation.
- Documentation: Record your reasoning, the classification, the authority gap, and the alternatives considered.
Show full worked solution
Full recommendation: Do not proceed with the public AI system. The proposed use fails the purpose limitation test and lacks the required authority. Recommended safeguards: (a) consult the university’s privacy office to confirm whether the feedback can be used in an AI system; (b) if approved, use only an institutionally approved AI system with appropriate data protection; (c) if no approved system exists, use a non-AI qualitative analysis method. Document all decisions.
Why this is appropriate: This decision protects student privacy, respects the university’s information governance, and avoids a potential incident. It also demonstrates responsible stewardship of information that belongs to the university and relates to students.
14. Guided Learning Activities
- Purpose: To connect FAIRER concepts (from Units 6–7) to information governance.
- Learning outcome: Identify stakeholders, information, accessibility needs, and contestability needs.
- Estimated time: 15–20 minutes.
Scenario: A non-profit organisation uses an AI system to screen job applications for a new program. The system is trained on past hiring data. The organisation is considering using a public AI system to summarise applicant cover letters and rank them. The information includes names, addresses, education, work history, and a “personal statement” section where applicants may disclose accessibility needs, cultural background, or family circumstances.
Task: Using the FAIRER lens, identify:
- Affected stakeholders
- Information involved and its classification
- Accessibility needs
- One possible unequal information risk
- Human authority
- One contestability need
Show model analysis
- Stakeholders: Job applicants (primary), hiring staff, the organisation, and potentially the AI provider.
- Information: Names, addresses, education, work history, personal statements. Classification: Personal, Sensitive personal (if accessibility or cultural info is disclosed), Confidential (hiring process), Internal (organisation use).
- Accessibility needs: Applicants with disabilities may rely on the personal statement to explain gaps or request accommodations. The AI may not handle this fairly.
- Unequal information risk: Applicants from marginalised backgrounds may be less likely to “tune” their cover letters for AI screening, or the AI may reflect past biases.
- Human authority: The organisation owns the process; hiring managers must make final decisions; applicants have rights to contest decisions.
- Contestability need: Applicants should be able to challenge an AI-generated ranking or summary, especially if it affects their eligibility.
Connection to this lesson: The organisation must classify the information, establish authority to use it in an AI system, and apply purpose limitation (hiring is the purpose; training or analytics may be secondary).
- Purpose: To distinguish the four concepts.
- Learning outcome: Distinguish privacy, confidentiality, security, and data protection.
- Estimated time: 10–15 minutes.
Instructions: For each scenario, identify the primary concept(s) at play and explain why.
- A hospital encrypts patient records to prevent unauthorised access.
- A student asks their professor to keep their grade private.
- An organisation has a policy that employee data will be deleted after 7 years.
- A journalist receives a leaked document and must decide whether to publish it.
Show model answers
- Security (technical protection). Also confidentiality (the encryption ensures only authorised access).
- Privacy (the student’s right to control information about their performance). Confidentiality (the professor has a duty to keep grades private).
- Data protection (lifecycle governance—retention and deletion policy).
- Confidentiality (the document is not publicly authorised) and data protection (the journalist must consider the source, accuracy, and consequences).
- Purpose: To practise provisional classification.
- Learning outcome: Classify information provisionally.
- Estimated time: 10–15 minutes.
Instructions: Classify each item as one or more of: public, internal, personal, sensitive personal, confidential, proprietary, privileged, security-sensitive, culturally governed, or insufficient information.
- A university’s annual report, published on its website.
- An employee’s home address stored in the HR system.
- A draft contract with a vendor, marked “negotiation draft – confidential.”
- A list of emergency evacuation routes for a government building.
- An Indigenous community’s oral history, shared with a researcher under community protocol.
Show model classifications
- Public (published).
- Personal, confidential (HR records).
- Confidential, proprietary (commercial negotiation).
- Internal, security-sensitive (could enable unauthorised access).
- Culturally governed, confidential (community protocol). Also personal if it contains identifiable information about community members.
Note: These are provisional. Final classification depends on the relevant authority and policy.
- Purpose: To identify roles and responsibilities.
- Learning outcome: Identify information owners, stewards, and authorised users.
- Estimated time: 10–15 minutes.
Scenario: A research team at a university has collected survey data from 500 participants about their mental health. The data is stored on a secure university server. The principal investigator (PI) is the lead researcher. A graduate student (you) is helping with analysis. The university’s research ethics board (REB) approved the study for a specific purpose: “understanding mental health service utilisation.”
Task: Identify:
- Information subject
- Owner or governing authority
- Steward or custodian
- Authorised user
- Reviewer
- Decision authority
- Incident authority
- Stop authority
Show model map
- Information subject: The 500 survey participants.
- Owner/governing authority: The university (through the research ethics board and the PI).
- Steward/custodian: The PI or the university’s research data manager.
- Authorised user: Research team members (including you) with appropriate training and approvals.
- Reviewer: The REB (for ethics) and the PI (for research integrity).
- Decision authority: The PI, with advice from the REB for changes to the research.
- Incident authority: The PI and the university’s privacy or research integrity office.
- Stop authority: The PI, the REB, or any team member who identifies a serious violation.
Application: You, as a graduate student, are an authorised user for analysis. You do not have authority to use the data for a new purpose (e.g., training an AI model) without REB approval.
- Purpose: To practise applying purpose limitation.
- Learning outcome: Apply purpose limitation to a scenario.
- Estimated time: 10–15 minutes.
Scenario: A public health agency collects anonymous data on flu vaccine uptake. The data is collected for “public health surveillance and response.” The agency is considering using the data to train a public AI chatbot that answers questions about flu vaccination. The chatbot would be free and publicly accessible.
Task: Apply the purpose-limitation test. Is the proposed use compatible? Why or why not? What would be required to make it compatible?
Show analysis
Original purpose: Public health surveillance and response (monitoring vaccine uptake, identifying outbreaks, planning interventions).
Proposed AI purpose: Training a public-facing chatbot to answer questions about flu vaccination.
Compatibility: The substantive goal (improving vaccination knowledge) may align with public health, but the method (using the data to train a chatbot) is a secondary use. The data was collected for surveillance, not for AI training. The agency would need to: (a) determine if the data is truly anonymous (if it can be re-identified, it is personal), (b) obtain a legal or policy basis for secondary use, (c) ensure the chatbot does not reveal sensitive information, and (d) consider whether the chatbot could cause harm (e.g., providing inaccurate advice). Result: Not compatible without additional safeguards and authorisation.
15. Exercises
- Difficulty: Introductory
- Estimated time: ~10 min
- Learning outcome: Distinguish privacy, confidentiality, security, and data protection.
Scenario: A company has a strict policy that all customer data must be encrypted and only accessible to authorised staff. The company also promises customers that their data will be used only for order fulfilment and will not be sold to third parties. A staff member accidentally sends a customer’s order details to the wrong email address.
Task: Identify which aspects relate to privacy, confidentiality, security, and data protection. Explain how they overlap.
Show sample answer
- Privacy: The company’s promise to use data only for order fulfilment and not to sell it respects customers’ control over their information.
- Confidentiality: The policy that only authorised staff can access customer data is a confidentiality obligation.
- Security: Encryption is a security control (protecting confidentiality and integrity).
- Data protection: The whole system—policy, encryption, access controls, and the incident response—is data protection.
- Overlap: The accidental email is a confidentiality breach (unauthorised disclosure), a security failure (human error, perhaps lacking safeguards), a privacy concern (customer information was exposed), and a data protection incident (the lifecycle governance failed).
- Difficulty: Introductory
- Estimated time: ~10 min
- Learning outcome: Classify information provisionally.
Scenario: A local school board has the following information:
- Student attendance records (with names and dates).
- A publicly available school calendar.
- Teacher performance evaluations (confidential, for HR use).
- A draft budget for the next school year (marked “internal – not for public release”).
- An Indigenous language curriculum developed in partnership with a local First Nation, governed by a community protocol.
Task: Classify each item using the nine provisional categories. Indicate where categories overlap or where you need more information.
Show sample classification
- Student attendance records: Personal, internal (school use), possibly confidential (if policy restricts disclosure).
- School calendar: Public.
- Teacher performance evaluations: Confidential, personal (about teachers), internal (HR use).
- Draft budget: Internal, confidential (marked not for release), possibly proprietary (if it reflects strategic planning).
- Indigenous language curriculum: Culturally governed (community protocol), confidential (may not be publicly shared without permission).
- Difficulty: Intermediate
- Estimated time: ~10 min
- Learning outcome: Identify information owners, stewards, and authorised users.
Scenario: You work as an administrative assistant in a university department. You have access to the faculty personnel files, which contain employment contracts, performance reviews, and personal contact information. A faculty member asks you to use a public AI system to “summarise their CV for a grant application.”
Task: Identify the information owner, steward, authorised users, and decision authority. Can you use the public AI system? Why or why not?
Show sample answer
- Owner: The university (the employer).
- Steward: The human resources department or records office.
- Authorised users: HR staff, department heads with a need to know, and you (as an administrative assistant) for administrative purposes.
- Decision authority: The faculty member’s supervisor, the HR director, or the university’s privacy office.
- Can you use the public AI system? No. You do not have authority to copy personnel files into a public AI system. The purpose (summarising a CV for a grant application) may be legitimate, but the method is not authorised. You would need to (a) get the faculty member’s consent, (b) check university policy, (c) use an approved system, and (d) ensure the information is not exposed.
- Difficulty: Intermediate
- Estimated time: ~12 min
- Learning outcome: Apply purpose limitation to a synthetic case.
Scenario: A city collects traffic camera footage for traffic management and safety (e.g., detecting congestion, identifying accidents). The footage is not publicly released. A city planner proposes using the footage to train an AI model that predicts pedestrian traffic patterns for urban planning. The planner argues that urban planning is a “related purpose.”
Task: Apply the purpose-limitation test. Is the proposed use compatible? What additional considerations apply?
Show sample answer
Original purpose: Traffic management and safety (real-time monitoring, incident response).
Proposed purpose: Training an AI model for pedestrian traffic prediction (urban planning).
Compatibility: The purposes are related (both involve transportation), but they are not the same. Training an AI model is a secondary use that may not have been contemplated when the footage was collected. Additional considerations:
- The footage may contain identifiable people (pedestrians, drivers), raising privacy concerns.
- The city may need a legal basis for secondary use (e.g., consent, a new policy, or de-identification).
- The AI model could introduce new risks (e.g., biases, errors) that the original purpose did not address.
- The city should conduct a privacy impact assessment and consult with the privacy office.
Result: Not compatible without additional authorisation and safeguards.
- Difficulty: Advanced
- Estimated time: ~12 min
- Learning outcome: Recommend whether information may be supplied, generalised, replaced, or excluded.
Scenario: You are a student working on a group project. Your group has collected interview transcripts from five participants about their experience with online learning. The transcripts contain names, student IDs, and detailed personal stories. Your group wants to use a public AI system to identify common themes in the transcripts for your final report.
Task: Should you supply the transcripts? If not, what alternatives exist? Make a recommendation and justify it.
Show sample recommendation
Recommendation: Do not supply the original transcripts to a public AI system. The information is personal and confidential (interviews with identifiable participants). You likely lack authority to share it with a third-party system, and the system’s terms may permit secondary use.
Alternatives:
- Use synthetic data: Create a set of synthetic interview excerpts that represent the themes without using real information.
- Redact: Remove names, IDs, and any direct identifiers. But remember—removing names may not anonymise the text (Lesson 8.2).
- Use an approved system: If your institution has an approved AI system with appropriate data protection, use that instead.
- Non-AI analysis: Manually review the transcripts and identify themes using a structured coding approach.
- Seek consent: If you have ethics approval and participant consent, you may be able to use the data under specific conditions.
Justification: The public AI system is not approved for personal/confidential information. The risk of exposure or secondary use is too high. The alternatives preserve the learning objectives while protecting participants.
16. Homework Problems
- Estimated time: ~8 min
Scenario: A person shares a personal story with a friend and asks them not to tell anyone. The friend agrees. Later, the friend shares the story with a third person.
Task: Identify the privacy and confidentiality aspects. What went wrong?
Show sample answer
Privacy: The person has a privacy interest in the story—they have a right to control who knows it.
Confidentiality: The friend accepted a confidentiality obligation when they agreed not to tell anyone.
What went wrong: The friend breached confidentiality (the obligation), which also violated the person’s privacy (the interest). The friend may have had access to the story, but did not have authority to share it.
- Estimated time: ~8 min
Scenario: A company has strong encryption and firewalls (excellent security). However, the company collects customer data without a clear purpose and shares it with third-party advertisers without consent.
Task: Does good security mean good privacy? Explain.
Show sample answer
No. Security is a means of protecting privacy, but it does not guarantee it. The company has strong technical security, but it violates privacy by collecting data without a legitimate purpose and sharing it without consent. Privacy also requires governance, transparency, and purpose limitation—not just encryption.
- Estimated time: ~10 min
Scenario: A hospital has the following records:
- Patient medical histories (with names, diagnoses, treatments).
- A public brochure about healthy eating.
- Internal policy documents on infection control.
- A list of hospital employees’ salaries (for payroll).
Task: Classify each using the nine categories. Indicate overlaps.
Show sample classification
- Patient medical histories: Personal, sensitive personal, confidential (doctor–patient), internal (hospital use).
- Public brochure: Public.
- Infection control policies: Internal, confidential (if not for public release), possibly security-sensitive (if it includes vulnerabilities).
- Employee salaries: Personal, confidential (HR records), internal.
- Estimated time: ~8 min
Scenario: A legal firm has a document that contains:
- Client names and contact details.
- Legal strategy notes (marked “attorney work product – privileged”).
- Financial projections for a client’s business (marked “confidential – proprietary”).
Task: Which categories apply? How do they overlap?
Show sample answer
- Personal (client names).
- Confidential (client information).
- Privileged (legal strategy).
- Proprietary (financial projections).
Overlap: The document is simultaneously confidential, privileged, and proprietary, and contains personal information. Each category imposes different obligations and protections.
- Estimated time: ~10 min
Scenario: A government department has a database of citizen complaints about public services. The data is collected for service improvement. A policy analyst (you) wants to use the data to train a public AI chatbot that helps citizens submit complaints more easily.
Task: Who owns the data? Who stewards it? Who has authority to approve this new use? What steps must you take?
Show sample answer
- Owner: The government (the department).
- Steward: The department’s records manager or privacy office.
- Decision authority: The department head, with advice from the privacy office and legal counsel.
- Steps: (1) Document the proposed use and its benefits and risks. (2) Consult the privacy office to determine if there is a legal basis for secondary use. (3) Conduct a privacy impact assessment. (4) Obtain formal approval. (5) Implement safeguards (e.g., de-identification, access controls). (6) Monitor and review.
- Estimated time: ~10 min
Scenario: A research participant signs a consent form that says: “I consent to the use of my data for this research study.” The research team later uses the data to train an AI model that will be sold to a commercial company for personalised advertising.
Task: Is this use covered by the consent? Why or why not? What principles apply?
Show sample answer
No. The consent is for this research study, not for AI training or commercial use. Even if the participant agreed to “research,” the new use is secondary and commercial—it is not compatible with the original purpose. Principles:
- Consent is purpose-specific—it does not authorise unlimited uses.
- Consent must be informed—the participant did not know about AI training or commercial sale.
- Consent may be withdrawn—the participant should be able to opt out.
- Secondary use requires new consent or a different legal basis.
- Estimated time: ~10 min
Scenario: A public library collects borrowing records to manage loans and recalls. The records include book titles, dates, and library card numbers (not names, but linked to a unique patron ID). The library wants to use the data to develop a “personalised reading recommendation” AI system that suggests books based on borrowing history.
Task: Is this use compatible with the original purpose? What are the risks? What safeguards are needed?
Show sample answer
Compatibility: The purposes are related (both involve library services), but they are different. Loan management is administrative; recommendations are a new service that may not have been part of the original collection purpose.
Risks:
- Reading history is sensitive—it may reveal political, religious, or personal interests.
- Profiling could lead to embarrassment or discrimination.
- The AI system may make errors or suggest inappropriate books.
Safeguards:
- Obtain opt-in consent from patrons for the recommendation service.
- Use anonymous or aggregated data if possible.
- Ensure the AI system does not share data with third parties.
- Provide a way for patrons to opt out or delete their history.
- Conduct a privacy impact assessment.
- Estimated time: ~8 min
Scenario: An employer collects employee performance data for performance reviews and promotions. The HR department proposes using the same data to train an AI system that predicts employee turnover.
Task: Is this a secondary use? If so, what additional considerations apply?
Show sample answer
Yes, this is a secondary use. The original purpose is performance review and promotion. The new purpose is turnover prediction. Considerations:
- Employees may not have consented to their data being used for turnover prediction.
- The AI model could introduce bias (e.g., predicting higher turnover for certain groups).
- The employer must comply with employment and privacy laws.
- If the model is used for decisions (e.g., who gets retention bonuses), it has a direct impact on employees.
- The employer should consult with employees, conduct a privacy impact assessment, and ensure transparency.
- Estimated time: ~10 min
Scenario: A research institution collects health data for a study on diabetes management. The data is stored on a secure server. A researcher wants to share a subset of the data with a commercial AI company to develop a predictive model for diabetes complications.
Task: What are the transfer risks? What protections are needed?
Show sample answer
Risks:
- Loss of control over the data once it is transferred.
- The commercial company may use the data for purposes beyond the research (secondary use).
- The company may have weaker security or privacy practices.
- Participants may not have consented to data sharing with a commercial entity.
- Legal or regulatory requirements (e.g., privacy laws) may be violated.
Protections:
- Obtain informed consent for data sharing, or have a legal basis.
- Use a data transfer agreement that restricts use to the specified purpose.
- Ensure the company has certified security and privacy practices.
- De-identify the data before transfer (but be aware that de-identification is not guaranteed).
- Conduct a privacy impact assessment and consult with the ethics board.
- Monitor the company’s use and audit periodically.
- Estimated time: ~8 min
Scenario: An organisation has strong firewalls and encryption (security), but no policy on how long to keep data, no process for handling access requests, and no incident-response plan.
Task: Is the organisation’s data protection adequate? Why or why not?
Show sample answer
No. Security is only one part of data protection. The organisation lacks governance (retention policy), process (access requests), and accountability (incident response). Data protection requires a complete lifecycle approach—not just technical controls. Without policies and processes, the organisation cannot demonstrate responsible use.
- Estimated time: ~10 min
Scenario: A university collects student survey data for institutional research and planning. The university wants to share the data with a government agency for national education policy analysis. The data is anonymised (or so the university believes).
Task: Is the purpose compatible? What about the anonymisation? What else should the university consider?
Show sample answer
Compatibility: The purposes are related (education research), but they are different. Institutional research is for the university’s own planning; national policy analysis is a broader, external purpose. This may be a secondary use requiring new authority.
Anonymisation: The university believes the data is anonymised, but they must verify that re-identification is not possible. Even if names are removed, indirect identifiers (e.g., department, year, course combination) may enable re-identification.
Additional considerations:
- Did students consent to their data being shared with the government?
- Is there a legal basis (e.g., a data-sharing agreement)?
- What security and privacy practices does the government agency have?
- What will the data be used for beyond the stated purpose?
Recommendation: Do not share without a formal data-sharing agreement, a clear legal basis, and verification of anonymisation. Consult the privacy office and the ethics board.
17. Mini-Project: Information Authority and Purpose Assessment
- Purpose: To synthesise the concepts from this lesson by assessing a bounded synthetic AI-supported activity.
- Learning outcomes: Classify information, identify authority, apply purpose limitation, and make a preliminary recommendation.
- Estimated effort: 75–100 minutes.
Context
Select one of the following bounded synthetic AI-supported activities. You will develop a structured assessment based on the template below.
- Option A (Education): A professor wants to use a public AI system to generate personalised feedback for student essays. The professor has access to the essays (which contain student names and ID numbers) and wants to upload them to the AI system.
- Option B (Professional practice): A social worker wants to use an AI system to summarise client case notes for a supervision meeting. The case notes contain sensitive personal information about clients (health, family, legal issues). The social worker has access to the notes as part of their job.
- Option C (Organisational): A marketing manager wants to use a public AI system to analyse customer feedback from product reviews. The reviews are stored in a company database and include customer names, product details, and ratings.
- Option D (Everyday life): A person wants to use a public AI system to organise their personal medical records (lab results, appointment notes) into a timeline for a new doctor. The records contain their name, date of birth, and health information.
Deliverable
Using the template below, develop a structured assessment. You may use the supplied-output option (no live system needed).
Template
- Human purpose
- Proposed AI use
- Information inventory
- Provisional classifications
- Information subjects
- Information owner or governing authority
- Steward or custodian
- Authorised users
- Original purpose
- Proposed purpose
- Compatibility assessment
- Secondary-use concern
- Required authorisation
- Information that must be excluded
- Safer alternative
- Escalation condition
- Stop condition
- Preliminary recommendation
Assessment criteria
- Completeness: All 18 items are addressed.
- Accuracy: Classifications and authority identifications are reasonable.
- Reasoning: The compatibility assessment and recommendation are justified.
- Privacy and security: You identify risks and propose safeguards.
- Realism: Your assessment reflects real-world constraints and responsibilities.
Self-review checklist
- I have classified all information provisionally.
- I have identified the owner, steward, and authorised users.
- I have compared the original and proposed purposes.
- I have identified secondary-use concerns.
- I have recommended a safer alternative if needed.
- I have identified escalation and stop conditions.
- My recommendation is clear and justified.
Show Hint 1: Getting started
Begin by listing all the information involved in the activity. Don’t just list “essays” or “case notes”—break it down: names, IDs, content, dates, etc. Then classify each item provisionally.
Show Hint 2: Authority
Think about who would have legitimate authority to use the information for the proposed AI purpose. Is it the individual, the organisation, or someone else? In many cases, you will need to go beyond “I have access” and identify a specific policy, consent, or legal basis.
Show Hint 3: Purpose limitation
Be honest: is the proposed AI use compatible with the original purpose? If there is any doubt, you should identify it as a secondary use and consider what additional authorisation would be needed.
Show solution architecture
- Human purpose: State the legitimate goal.
- Proposed AI use: Describe the AI system and what it will do.
- Information inventory: List all data fields and artefacts.
- Provisional classifications: Assign one or more categories to each item.
- Information subjects: Identify the people the information is about.
- Owner/authority: Identify the organisation or person with ultimate authority.
- Steward/custodian: Identify who manages the information.
- Authorised users: Who is allowed to access it for the original purpose?
- Original purpose: Document the original reason for collection.
- Proposed purpose: Document the AI-supported use.
- Compatibility: Analyse whether the purposes are compatible.
- Secondary use: Identify any secondary uses.
- Required authorisation: State what additional permissions are needed.
- Information to exclude: Identify fields that should not be supplied.
- Safer alternative: Propose a safer approach.
- Escalation: When should you escalate to a higher authority?
- Stop condition: When should you not proceed?
- Recommendation: State your final recommendation clearly.
Show illustrative partial example (different context)
Example context: A course coordinator wants to use an AI system to generate quiz questions from lecture slides. The slides are stored on a university learning management system and are marked “Instructor use only – not for redistribution.”
Partial assessment:
- Human purpose: Create additional assessment materials for students.
- Proposed AI use: Upload lecture slides to a public AI system to generate multiple-choice questions.
- Information inventory: Lecture slides (text, images, examples).
- Provisional classifications: Internal (university use), confidential (marked instructor use only), proprietary (instructor’s intellectual property).
- Information subjects: The instructor (as the author), the university (as the employer).
- Owner: The instructor (as creator) and the university (as employer).
- Steward: The university’s learning technology office.
- Authorised users: The instructor and teaching assistants for teaching purposes.
- Original purpose: Teaching and learning.
- Proposed purpose: Generating quiz questions.
- Compatibility: The purpose is compatible (both teaching), but the method (public AI) may not be authorised. The slides are marked “not for redistribution.”
- Secondary use: The AI provider may use the input for model improvement.
- Required authorisation: Permission from the instructor and the university’s privacy office.
- Information to exclude: Any examples or images that identify people or sensitive topics.
- Safer alternative: Use an approved AI system or generate questions manually.
- Stop condition: If the university policy prohibits using public AI for course materials.
- Recommendation: Proceed with safeguards: use an approved system, remove any sensitive content, and obtain instructor permission.
Submit your completed template as part of your Unit 8 portfolio. This is formative—it prepares you for Assignment 3.
18. Self-Check Quiz
Question 1: Which of the following best describes confidentiality?
Show answer and explanation
Answer: c.
Explanation: Confidentiality is about the duty or obligation to protect information. Privacy is the person’s right (a), security is technical measures (b), and data protection is the full system (d).
Question 2: A document marked “DRAFT – INTERNAL USE ONLY” should be classified as (select all that apply):
Show answer and explanation
Answer: b and c.
Explanation: “Internal use only” indicates it is for internal use (b) and restricts disclosure (confidential, c). It is not public (a), and unless it contains trade secrets or intellectual property, it may not be proprietary (d). However, in some contexts, it could also be proprietary. This illustrates that classification is context-dependent.
Question 3: True or false: If you have access to information, you have authority to use it in any AI system.
Show answer and explanation
Answer: False.
Explanation: Access is a technical permission. Authority is a legal, policy, or ethical permission. You may have access for one purpose but not for another.
Question 4: Which of the following is the best example of purpose limitation?
Show answer and explanation
Answer: c.
Explanation: Purpose limitation restricts use to the authorised purpose. (a) is security, (b) is a violation of purpose limitation, and (d) is retention.
Question 5: A research team wants to use survey data (originally collected for academic research) to train a commercial AI model for a startup. This is an example of:
Show answer and explanation
Answer: b.
Explanation: Using data for a purpose beyond the original authorised purpose (academic research → commercial AI training) is secondary use. It may or may not be permitted; it requires new authorisation.
Question 6: You are a student and you have access to your university’s internal course evaluation data. You want to use it to write a blog post about teaching quality. You will remove student names. Should you proceed?
Show answer and explanation
Answer: c.
Explanation: Access does not create authority. The data is internal and confidential. Removing names may not anonymise it (see Lesson 8.2). Blogging is not a permitted purpose. You should seek authorisation or use a non-AI alternative.
Question 7: Which of the following is the most appropriate response if you are unsure whether you have authority to use information in an AI system?
Show answer and explanation
Answer: c.
Explanation: When unsure, stop and seek guidance. Proceeding without authority risks privacy and security incidents. Deleting after use does not undo the exposure.
19. Assignment Connections
Lesson 8.1 provides the foundation for the information-governance portion of Assignment 3. Retain the following evidence:
- Your information inventory and provisional classifications.
- Your authority map (owners, stewards, authorised users, decision authorities).
- Your purpose-limitation analysis (original vs. proposed purpose, compatibility, secondary use).
- Your preliminary recommendation and justification.
- Any escalation or stop conditions you identified.
PROTECT elements to apply in Assignment 3: P (Purpose and authority), R (Review and classification), O (Ownership and stewardship). Remaining elements will be covered in Lessons 8.2–8.4.
Reflection prompts:
- How do privacy, confidentiality, and security differ, and why does the distinction matter for your professional practice?
- What is one example from your experience where you had access to information but may not have had authority to use it?
- How does purpose limitation affect the way you think about using AI in your work or studies?
- What types of information would you never enter into a public AI system, and why?
Do not share real personal or confidential information in your journal.
Show reflection-planning checklist
- I have considered how privacy, confidentiality, and security differ.
- I have reflected on a situation involving access and authority.
- I have considered how purpose limitation affects my AI use.
- I have identified information I would not enter into a public AI system.
20. Key Takeaways
- Privacy is about people’s control over information; confidentiality is an obligation to protect it; security is the technical means; data protection is the full governance framework.
- Access does not equal authority. You may be able to view information, but that does not mean you are authorised to use it in an AI system.
- Information classification is provisional. Use the nine categories as a starting point, but verify the final classification with the governing authority.
- Purpose limitation requires that information be used only for the authorised purpose. A new purpose requires new authorisation.
- Secondary use (e.g., AI training, analytics) is a different purpose and may not be permitted without consent or a legal basis.
- When in doubt, stop. If you lack authority, if the purpose is incompatible, or if you are unsure, do not supply the information.
- PROTECT begins with Purpose and Authority (P), Review and Classification (R), and Ownership and Stewardship (O)—all covered in this lesson.
21. Key Terms
- Privacy
- Appropriate control, governance, use, access, disclosure, and protection of information about people.
- Confidentiality
- An obligation or expectation that information will be accessed, used, and disclosed only as authorised.
- Security
- Protecting information and systems against unauthorised access, disclosure, alteration, loss, destruction, misuse, and unavailability.
- Data protection
- The policies, legal requirements, governance processes, and controls used to manage information responsibly throughout its lifecycle.
- Information classification
- The provisional assignment of information to categories such as public, internal, personal, confidential, proprietary, privileged, security-sensitive, or culturally governed.
- Authority
- The legitimate right to access, use, disclose, or make decisions about information.
- Consent
- Permission given for a specific purpose; it is purpose-limited and may be withdrawn.
- Purpose limitation
- The principle that information may be used only for the authorised and specified purpose for which it was collected.
- Secondary use
- Using information for a purpose beyond the original authorised purpose.
- Information subject
- The person to whom the information relates.
- Information owner
- The entity with ultimate legal or policy authority over information.
- Information steward
- The person or office responsible for managing information on behalf of the owner.
- Authorised user
- A person with legitimate permission to access or use information for a defined purpose.
- Direct identifier
- Information that uniquely identifies a person (e.g., name, ID number).
- Indirect identifier
- Information that can be combined with other data to identify a person (e.g., postal code, date of birth).
- Incident
- An event involving unauthorised access, disclosure, alteration, loss, or misuse of information.
22. Looking Ahead: Lesson 8.2
In Lesson 8.1, you established the foundation for information governance. In Lesson 8.2: Data Minimization, De-Identification, Memory, and Retention, you will learn to:
- Apply data minimisation to prompts and workflows.
- Distinguish deletion, redaction, pseudonymisation, anonymisation, aggregation, and synthetic substitution.
- Explain why removing names may not anonymise information.
- Recognise re-identification and inference risks.
- Evaluate conversation history, persistent memory, retention, and secondary use.
This builds directly on the classification, authority, and purpose-limitation foundation you have established here.
23. Sources & Further Reading
- Primary: Chapter 10: Privacy, Security, and Data Protection (COMP 3270 course text).
- Supporting: Chapter 6: Human-AI Collaboration (authority, accountability).
- Supporting: Chapter 7: Hallucination, Error, and Uncertainty (risk awareness).
- Supporting: Chapter 8: Verification, Evidence, and Epistemic Responsibility (evidence for authority).
- Appendix A: Generative AI Glossary.
- Appendix D: Responsible-Use Toolkit.
- Appendix G: Discipline-Specific Checklists.
Some textbook content, definitions, examples, and figures are marked with placeholders. These will be supplied by the course coordinator or subject-matter expert.