BETA This site is in beta. Information is still being added and reviewed.
Training hub Module 4

Privacy and immigration-status protection

What clients should never type into any AI tool, how LINC handles data, and how to give honest reassurance that is neither falsely comforting nor needlessly frightening.

Learning objectives

By the end of this module, the CHW can:

  1. List what a client should never enter into an AI tool.
  2. Explain, in plain words, how LINC handles what a client types or uploads.
  3. Coach a client to cover names and ID numbers before sharing a document photo.
  4. Give calibrated reassurance: honest about protections and honest about limits.

Facilitator outline

0:00–5:00
Recap and the stakes

For clients living with immigration precarity, privacy is not abstract. Name that plainly and set a calm, respectful tone.

5:00–15:00
What never to enter

Immigration status, A-numbers and case numbers, Social Security numbers, full legal name tied to sensitive details, and exact home address. Ask general questions instead of ones that identify the person.

15:00–25:00
How LINC handles data

Walk the transparency guide: what LINC does and does not do with a message or an uploaded photo. Be precise, and do not promise more than the guide states.

25:00–33:00
Practical rules for document photos

Cover names and ID numbers before photographing an intake form or insurance letter. Show how to crop or place a finger over identifiers. Ask the AI only what the client needs to understand.

33:00–45:00
Activity: the client who fears data sharing

Run the role-play below. Practice honest, calibrated reassurance and concrete protective steps. Close with a teach-back.

45 min
Add 5 minutes of Q and A on your site's specific data policy to reach 50 minutes.

Key teaching points

Start from respect for the client's caution. Many clients are right to be careful about where their information goes, and the job here is to channel that caution into specific, workable rules rather than to talk them out of it. The core rule is simple: do not enter anything that identifies you or your status. Ask "how do I find a therapist who takes Medicaid" rather than typing a name, a case number, or a status. When a client needs help reading a document, a photo can be useful, but names and ID numbers should be covered first.

LINC chat
MEDICAID
What does this letter mean, and is there anything I need to do?
It looks like a Medicaid renewal notice. Here is the key information from what you sent:
What it is
Medicaid renewal notice
Deadline
Return the form by March 14
Do this
Complete and return the renewal form to keep your coverage
Call
The county number printed on your own letter
The name and ID number were covered before the photo was taken, so the tool never receives them. Notice the last row: the product points the client back to the number on their own letter instead of inventing one, and the deadline is flagged in vermilion because it is time-sensitive.

When you describe how LINC handles data, stay exact. Use the transparency guide's wording and do not improvise stronger promises. Calibrated reassurance sounds like this: "LINC does not save your photo, does not use it to train AI, and does not share it with immigration authorities. Even so, cover your name and any ID numbers first, because that is the safest habit anywhere." That is honest about the protection and honest about the limit. Avoid both failure modes: false comfort that oversells safety, and vague fear that leaves the client too scared to use a tool that could help them reach care.

Why this matters (evidence)

Onboarding that is upfront about a tool's limits is an evidence-backed way to prevent people from over-relying on it (Passi and Vorvoreanu, 2022). Being specific about what LINC does and does not do with data serves the same purpose: it builds trust that is grounded rather than blind, which is exactly the calibrated trust taught in Module 2.

Machine translation can quietly drop safety content. When an AI "simplified" a Korean consent form, readability improved but the quality of risk disclosure fell significantly (Oh and colleagues, 2025). After any translation, a person confirms the warnings survived.

Privacy fear is rational. Immigration enforcement chills care even for people who are lawfully present (Herring and Barnow, 2025). Lead with a plain statement of the facts before asking anything.

Main activity

Role-play: the client who fears data sharing

Pair up. One CHW plays a client who is anxious that anything typed into the chatbot could reach immigration authorities; the other coaches. The client opens with:

Client"I don't want to type anything into that. What if it gets sent to immigration? I've heard they can pull all of this. Maybe I just shouldn't use it at all."

The coach's task is to acknowledge the concern as reasonable, state the actual protections in the transparency guide's words, and give two or three concrete rules the client controls: ask general questions, never enter status or ID numbers, cover identifiers on any photo. Good facilitation leaves the client both safer and still able to use the tool. Harmful moves to flag in debrief: promising total safety, or dismissing the fear so the client disengages entirely.

Materials

Facilitator self-check
  1. Did I use the transparency guide's exact wording instead of inventing stronger promises?
  2. Did the client leave with concrete rules they control, not just reassurance?
  3. Did I avoid both false comfort and needless fear?