Skip to main content
Vizain

Trust and research ethics

Trust is not a badge. It's a workflow.

Responsible research demands participant protection, methodological transparency, traceable evidence and researcher oversight, not just impressive output.

Principles in practice

The details matter.

How Vizain protects participants and keeps every conclusion open to inspection.

  • Participant protection

    Names and contact details stay in a separate encrypted vault. They never reach research data, AI prompts or reports, and consent is recorded for every study.

  • Traceable conclusions

    Every finding links to the passages, statistics and documents behind it, and is tested against the evidence rules you set.

  • Human-led intelligence

    AI drafts and suggests; researchers review and approve. Every AI step is logged, and numbers and rule checks are computed in code, not by the model.

  • Control and governance

    Roles and permissions, data residency, retention settings, audit logs and organisation-level administration.

Confidence comes from what you can inspect, not what you are asked to believe.

The full safeguards register, data regions, AI use and sub-processors are below.

See the safeguards

Safeguards register

Each safeguard has a reference so you can cite it in your own ethics and data protection documents.

Anonymisation

  • PV-01

    No direct identifiers in research data

    No research table, file, log, analytics event or AI prompt holds a participant's name, title, phone, email, address, ID number, face or kept voice recording.

  • PV-02

    Anonymisation gate

    Every answer, upload, transcript, document, photo and imported dataset is scanned and redacted before it is saved. Items that cannot be cleared automatically are held encrypted for at most 24 hours for a researcher to fix or discard.

  • PV-03

    De-identified transcripts

    Names in transcripts become role labels such as [community leader]; places finer than the project's granularity are generalised.

  • PV-04

    Sampling frames stay on your device

    Beneficiary, member or customer lists used for sampling are processed in the browser. Only pseudonymous IDs and stratification attributes are uploaded; the linking key stays with you.

  • PV-06

    No names or titles in instruments

    Questionnaires, guides and translations are scanned and cannot be published while they contain participants' names, job titles tied to individuals, phone numbers or emails.

  • PV-09

    Pseudonymous longitudinal linking

    Baseline–endline and tracker waves are linked with pseudonymous keys, never identifiers.

Contact vault

  • PV-05

    Separate encrypted contact vault

    Contact details needed for invitations, operational forms or member polls are held in a separate, field-level encrypted vault with per-organisation keys, never joined to research data, never sent to AI, and deleted automatically after fieldwork closes.

Location and images

  • PV-07

    Coarsened location and safe photos

    Precise GPS is used on the device for verification and stored only as an administrative area or grid cell. Photos showing faces or readable documents are blocked on the device.

Re-identification protection

  • PV-08

    Re-identification protection

    Small cells are suppressed, exports are checked for k-anonymity on quasi-identifiers, and quotes are screened for identifying detail.

Verification

  • PV-10

    Automated PII verification

    An automated PII scanner test suite runs on every data path, and a scheduled job scans stored data and alerts on any detected identifier.

Recordings and transcripts

  • PV-12

    Audio deleted after transcription; transcripts deleted at close

    Audio is deleted as soon as it is transcribed. Transcripts are deleted when the project is closed; published reports and anonymised quotes remain. Every deletion is logged.

Security

  • SC-01

    Encryption in transit and at rest

    TLS 1.2+ in transit; encryption at rest; contact-vault fields encrypted with per-organisation keys; field-app data encrypted on the device.

  • SC-03

    Signed, expiring links for files

    All media and documents are served through signed, expiring URLs; there are no public buckets.

  • SC-04

    Rate limiting and bot protection

    Public form and survey endpoints are rate limited and protected against bots.

  • US-02

    Database-enforced tenant isolation

    Row-level security in the database keeps every organisation's data separate.

  • US-04

    Immutable audit log

    Access to data, exports, research-setting changes and report publication are recorded in an append-only audit log.

Privacy

  • PR-01

    Data residency

    Each organisation's data stays in its chosen region.

  • LL-06

    No training on customer data

    Customer data is never used to train AI models; zero-retention provider settings are used where available, and the AI gateway refuses any payload that fails the PII scanner.

Research ethics

  • ET-04

    AI transparency for participants

    Participant-facing pages state where AI is used, how data is anonymised and how to contact a human.

Where data is stored

Each organisation chooses the region its data is stored in. Regions available:

  • London (eu-west-2)
  • Frankfurt (eu-central-1)
  • N. Virginia (us-east-1)
  • Singapore (ap-southeast-1)

How we use AI

AI helps with the work listed below. Each project publishes its own AI-use disclosure, participant-facing pages say where AI is used, and every report's methodology records which steps used AI and how outputs were checked.

  • Translating instruments and answers
  • Analysing anonymised open-text answers
  • Transcribing recordings, which are then deleted
  • Finding and removing identifiers
  • Checking data quality
  • Drafting reports for a person to edit
  • Customer data is never used to train AI models.
  • Every AI request passes through a gateway that refuses any payload containing a personal identifier.
  • Numbers and rule checks are calculated by code, not by a language model.

Sub-processors

Companies that process data on our behalf. The list is generated from the same configuration as our data processing agreement.

Sub-processors
CompanyPurposeLocationStatus
Amazon Web Services EMEA SARL / Amazon Web Services, Inc.Hosting: application servers, PostgreSQL database, file storage, key management (KMS) and backups.Organisation's data regionOnly if enabled
DigitalOcean, LLCHosting: application servers, PostgreSQL database, file storage (Spaces) and backups.United Kingdom (London)Only if enabled
Cloudflare, Inc.Domain name service and the website's anti-spam check (Turnstile).Global networkOnly if enabled
Anthropic, PBCAI model inference for the features listed in each project's AI disclosure (translation, analysis of anonymised open text, transcript de-identification, report drafting).United StatesIn use
Amazon Web Services (Simple Email Service)Sending survey invitations, reminders and data-subject request verification emails.Organisation's data regionOnly if enabled
Twilio Inc.Sending SMS and WhatsApp invitations and reminders.United StatesOnly if enabled
Sendinblue SAS (Brevo)Sending survey invitations, reminders, data-subject request verification and staff emails; optionally SMS.European Union (France)Only if enabled
Stripe Payments Europe, Ltd.Subscription billing and payment fields in operational forms.European Union / United StatesOnly if enabled

Certifications

The platform is designed to meet the controls of SOC 2 Type II and ISO 27001. We will list certifications here only once an independent auditor has issued them.

Security documents

Customers can generate a data processing agreement, sub-processor list and records of processing from their settings. For security questionnaires, contact us.