Trust & security

Student privacy is enforced by the system, not by policy alone.

Who can see what is decided in the data layer, on every request. This page describes how the platform is built. For your security review we'll document the production environment in full, including a completed HECVAT.

Access

  • Single sign-on. People sign in through your institution's identity provider (SAML or OIDC, such as Shibboleth, Entra ID or Okta). We never hold their passwords, and they sign in again at your campus at least every 12 hours.
  • Deny by default. Signing in proves who someone is; it doesn't grant access. Your IT office controls who can sign in, through your identity provider, and your IR office gives each person a role and scope, under whatever approval process your institution sets. Anyone else is turned away.
  • Roles and row scope. An IR analyst sees the whole institution; a dean their college; a chair their department; an advisor their current advisees; faculty the students in their own sections. Scope is applied to every result, list and export.
  • Sensitive fields by role. Directory, academic-record, sensitive and HR data each have their own permissions, so salaries or demographics are visible only to roles that need them.
  • Our staff have no access to your student data. The role we use to operate the platform has no data access at all.

Privacy protections

  • Small-cell suppression. Any group of fewer than 10 students is suppressed in aggregate results, together with the related cells that would let it be worked out by subtraction.
  • No attachments. Scheduled reports are delivered as a link to the owner's verified institutional email, never as an attachment, so access is checked again when the report is opened.
  • Labelled exports. Every PDF carries the definitions, the scope it reflects, who it was prepared for, and an ID that ties it to its audit record.
  • Only the data you choose to provide. If your institution doesn't share something, such as salaries or test scores, the features that need it are switched off for you.

Audit trail

Every sign-in, question, report, export, student list and IPEDS file is recorded: who, in which role, what was requested, and how many rows came back. Audit records never contain student values or the text of questions. Refused sign-ins are recorded too, without the person's email address.

The AI

The AI's only job is to turn a question into a request for a governed metric. It receives the question, the catalog of metric definitions, and your institution's lists of programs, colleges and departments (names and codes) so it can filter by them. It never receives student or staff records or results, which are computed afterwards by the governed engine under the user's role and scope.

We use Claude models from Anthropic, run inside Google Cloud's Vertex AI in our own cloud projects. Under Google Cloud's terms, your data isn't used to train or fine-tune models without your permission. For abuse monitoring, Google may keep prompts and responses for some models for up to 30 days. Your agreement with us states which model and settings your institution's questions use, or you can switch the AI off and pick metrics directly.

Data and infrastructure

  • Google Cloud. Hosting, data, sign-in and AI all run on Google Cloud, defined in code and reviewed before every change. Data is encrypted in transit and at rest.
  • A firewall in front. Every request passes Google Cloud Armor, a web application firewall, which rate-limits each address and filters common attacks before anything reaches the service. The service accepts no other route in.
  • Isolation. Each institution's data is kept in its own dataset. Cross-institution benchmarks use aggregates only, with small-cell suppression; student records never leave their home institution's dataset.
  • No real student data outside production. Development and testing use a fully synthetic university.
  • Your data stays yours. Our agreement covers how your data is returned or deleted when you leave.

Compliance

AreaStatus
FERPAWe act as a school official under your institution's direct control, set out in a data-processing agreement.
HECVATWe'll complete your institution's HECVAT as part of your review.
SOC 2Controls are designed in from the start (access in the data layer, audit logging, change review). An independent audit is planned; we'll share our roadmap.
AccessibilityWe design to WCAG 2.1 AA. A VPAT is in preparation. Accessibility statement

Subprocessors

ProviderPurpose
Google CloudHosting, data storage, single sign-on (Identity Platform) and AI model hosting (Vertex AI)
Anthropic, through Google Cloud Vertex AIThe Claude model that turns questions into metric requests
Google Workspace (Gmail)Sends scheduled-report notifications: a secure link and the report's name, never student data or attachments

Questions

For security questionnaires, our data-processing agreement, or to report a concern, email security@institutionalresearch.ai.