Data-responsibility checklist for AI tools

Use before entering any data into an AI tool. Aligns with IASC Operational Guidance on Data Responsibility (2023) and OCHA Data Responsibility Guidelines (January 2025).

Text equivalent: Plan, assess, collect, share, analyse, store, close. Apply at each stage before using AI on operational data. See visual aids.

Before you type

I know what data classification applies (public, internal, confidential, sensitive operational, personal).
I confirmed this tool is approved for this data type in my organization.
I am using only the minimum data needed for the task.
I removed names, contact details, case numbers, and precise locations where not required.
I checked whether the vendor may use my input for model training or share it with third parties.
I know where data is processed and stored (country/region).

Safeguarding and protection

I am not entering child protection case notes, GBV survivor records, or other highly sensitive protection information unless explicitly approved for this tool and role.
I know my organization’s safeguarding, PSEA, and protection focal points before handling disclosures or incidents.
If someone discloses harm in a meeting or chat, I will pause, avoid plenary detail, and refer to those focal points (not investigate myself).
I treat re-identification risk seriously: combined fields can identify people even without full names.

Primary sources: ICRC Handbook on Data Protection (3rd ed.); Core Humanitarian Standard (2024) (Commitment 5). Not legal advice.

During the task

I am not combining datasets in ways that could re-identify people.
I am not uploading beneficiary lists, protection case notes, or medical records unless explicitly approved.
I saved a record of what I entered and when (without copying sensitive content into unsecured notes).

After the output

I verified consequential claims against an authorized source.
I checked translations for dropped negations, wrong numbers, or changed meaning.
I named who will approve before any operational use.
I deleted unnecessary copies from the tool if the platform allows it.
I reported a data incident to my organization’s contact if sensitive data was entered by mistake.

Red flags: do not proceed

Red flag	Why it matters
Beneficiary phone numbers in a public AI chat	Re-identification and privacy breach
Shelter locations tied to named individuals	Safety risk
Donor financial details	Confidentiality and fraud risk
Unvetted vendor with unclear data retention	Loss of control over sensitive data
GBV or child protection details in unapproved tools	Severe harm and policy breach

Need help?

Escalate to your privacy lead, protection adviser, safeguarding or PSEA focal point, or IT security team. Amber and red decisions require organizational review, not individual judgment alone.

See Session 3 safeguarding module and facilitator guide for training-room escalation.

Download this page as an accessible PDF