May 6, 2026 · Scheman Building, Ames, IA

CIRAS   |   Iowa State University

AI Fluency / Cycle 03

Safeguard Information

Experiment: Useful Information, Appropriate Protection

Recognize sensitivity, trace exposure, prepare only what is needed, evaluate where it goes, and decide whether it can be used.

How to Use This Page

Watch the instructor make information-handling decisions using a fictional service record. Then practice with a different engineering-support case. AI can help evaluate options, but you make the decision.

15 min guided20 min independent5 min compare and reflect
Classroom safety: All records below are invented. Use these examples, not live customer information. The environment descriptions are hypothetical, not provider recommendations or proof of organizational approval.

Guided Experiment: Protect a Manual Lookup

15 minutes · Follow along with the instructor

The job: Prepare a brief for a later lookup of the right published technical manual. You are not asked to locate a manual or make a repair decision.

Fictional company record
FICTIONAL SERVICE RECORD | Case SVC-2148 | October 28, 2026
Source: Customer email, with internal notes added by the service team
Customer: Maya Ellis | maya.ellis@example.test | 515-555-0117
Location: 47 Test Ridge Road, Exampleton, Iowa
Product: ERV-240C | unit label: Revision R2 | installed about 2022
Customer report: "The amber indicator flashes twice after I replaced the filter. The unit still seems to run."
Evidence gap: Customer mentioned a nameplate photo but did not attach it.
Internal sales note: Dealer-specific purchase price $765; not approved for public release.
Internal engineering note: Draft memo ENG-12 describes an unapproved workaround. Do not share with customers.
TASK: Prepare a minimum-necessary brief for an approved later lookup of the published customer-facing product manual. Do not identify a manual, diagnose the equipment, or issue repair instructions.

0:00-2:00

1. Recognize: What needs protection?

Identify public product facts, internal notes, personal information, and proprietary details. Discuss how an ordinary unit ID can become sensitive when linked to a specific customer.

2:00-4:00

2. Trace: Where might it be exposed?

Original case>Working excerpt>AI tool>Output>Recipient>Retention

Where could information be revealed through uploads, connected files, generated answers, sharing, or storage? No model training does not answer questions about processing, access, or retention.

4:00-9:00

3. Prepare: Keep what the task needs

Compare the raw case with a minimized version. Would replacing the customer’s name with “Customer A” make it anonymous? Which technical facts would be lost by excessive removal?

Reveal two alternative working excerpts
Fit-for-purpose, minimized draft

TASK: Locate the applicable published customer-facing technical manual.
Product: ERV-240C | unit revision R2 | installed approximately 2022.
Reported issue (unverified): Amber indicator flashes twice following a filter replacement; unit reportedly still runs.
Open question: Nameplate photo mentioned but not supplied. Confirm product identification if needed.
OUT OF SCOPE: Troubleshooting, parts recommendations, unpublished engineering instructions.

Over-minimized draft

TASK: Find a manual.
A ventilation unit has a flashing light.

Neither version is automatically anonymous or approved for every environment. The second does not retain enough detail to identify a specific manual.

Compare the protection methods from the presentation
Method Example and limitation
Redact Remove the street address; preserve model and reported symptom.
Mask Show only the final phone digits when genuinely useful. Still potentially personal.
Pseudonymize Use Case A instead of a name. A lookup key may still reconnect the record to a person.
Generalize / aggregate Use approximate installation year or counts across cases for a trend question, if sufficient.
Sanitize Remove hidden comments, tracked changes, or metadata before sharing a Word/PDF excerpt.

Data states: Raw contains original details; pseudonymized data can remain linkable; genuinely anonymized data requires careful re-identification assessment; synthetic data is invented for practice.

9:00-12:00

4. Evaluate: Which environment is appropriate?

Use the comparison below. An environment label alone cannot answer whether the actual task is authorized.

12:00-15:00

5. Decide: What would you do?

Ask: Purpose? Minimum data? Permission? Environment and access? Remaining risk? Choose a defensible next action.

Proceed, if approvedModify the dataChange environmentPause and ask

Optional instructor prompt

Copy the fictional case into your AI tool, then copy this prompt. Treat any classification or recommendation as a proposal to verify.

Guided example prompt

Using the FICTIONAL service record I pasted, help me prepare information for one task: identifying the appropriate published customer-facing technical manual, NOT diagnosing the equipment.

1. Classify the main details as public, internal, personal, or confidential/proprietary. Categories may overlap.
2. Show the smallest useful working brief that preserves model, revision, reported symptom, and important uncertainty. Remove unrelated identifiers, pricing, and draft engineering instructions.
3. Explain in two short sentences what might still be sensitive and what information was deliberately left out.
Do not claim the result is anonymous or that any AI service is approved. Keep the entire response under 180 words.

Compare Processing Environments

These are hypothetical descriptions. Compare what is known, unknown, and who would need to approve the use.

Environment Example conditions What must still be checked?
Individual cloud account Personal account; organization has not approved it for customer records. Actual terms, processing, training settings, retention, access, and permission.
Company-managed AI Role-based access and written processing/retention terms are available. Whether this specific type of record and purpose is approved; output sharing and deletion.
Organization-controlled / private Infrastructure and access may be managed under company controls. Actual provider access, contracts, configuration, retention, and approved uses.
Locally operated model or tool Processing may remain on local hardware, without using a remote AI provider. Device security, users, logging, output storage, retention, and approval.

Remember: A vendor saying “not used for training” is not a complete assurance. Encryption, local deployment, and enterprise features also do not grant permission by themselves.

Independent Experiment: Protect an Engineering Escalation

20 minutes · Make the decisions yourself

Different job: Prepare information for internal engineering to evaluate a replacement-part question. This time, a complete unit serial may be needed, so deleting everything identifiable could make the task impossible.

Different fictional record
FICTIONAL INTERNAL EXCHANGE | Service case SVC-2260 | October 30, 2026
Source: Customer request, parts-sales note, and engineering reply
Customer: Morgan Tate | morgan.tate@example.test | 515-555-0146
Service address: 82 Sample Creek Lane, Exampleton, Iowa
Equipment: HRV-180B | revision B | serial HRV180B-20-04682
Customer question: Is replacement control board RB-180-B compatible? A technician says the connector shape differs from the original.
Parts-sales note (internal): An old catalog lists RB-180-A. Dealer price under contract Q-77 is $94. Do not send contract pricing outside the authorized channel.
Engineering reply (internal, not customer-approved): RB-180-C is a prototype, not an approved replacement. Full unit serial is required to check the correct configuration. No compatibility decision has been made.
TASK: Prepare a short internal escalation for engineering to investigate compatibility, NOT a response to the customer. Identify what information an approved internal workflow would need and what must not go to an unapproved AI service.

0:00-4:00

1. Recognize and trace

Classify what is personal, internal, proprietary, or contract-restricted. Identify two points where the information or output might be exposed.

4:00-11:00

2. Prepare a useful but limited escalation

Decide what engineering needs, what can be redacted or masked, and what must stay inside approved systems. Draft a short internal handoff. Do not invent a replacement-part answer.

11:00-16:00

3. Compare ways to handle it

Compare a minimized excerpt in an approved AI environment with keeping the full technical record in a restricted internal system. Consider which option preserves a needed full serial number and why a consumer chat may not be appropriate.

16:00-20:00

4. Make the decision

Choose proceed, modify, change environment, or pause for approval. Explain one remaining risk and who must authorize the actual processing.

Optional prompts: choose an approach

Copy the fictional case first, then choose one prompt.

Option A: Draft and critique

This is an entirely FICTIONAL manufacturer case. The task is to prepare an internal engineering escalation about control-board compatibility, not a customer reply.

Suggest a minimum-necessary information set. Separate (a) information engineering truly needs, (b) information that can be omitted or masked, and (c) information that must remain restricted. Recommend whether this could use an unapproved consumer AI tool, needs an approved organization-controlled environment, or should wait for authorization. Give one remaining risk. Keep the answer short; do not decide which board fits.
Option B: Make the decisions yourself

I will decide how to handle this fictional engineering-support record. Act as a reviewer, not the decision-maker. Ask me three short questions, one at a time, about (1) minimum information needed, (2) how the information could be transformed or separated, and (3) access, retention, and approval for the processing environment. Then challenge one weakness in my proposed approach. Do not assume that removing names makes data anonymous.

Compare and Reflect

5 minutes · Group discussion

  • Why could removing a customer’s name still leave a record identifiable?
  • When would removing more information make the work less accurate or useful?
  • What changes when an AI tool is company-managed, private, or local? What questions remain?
  • Which decision would you make if the data is needed but authorization is unclear?