Protect student data and research IP at
Education platforms and research institutions face the strictest student data laws and the tightest budgets in this series, where 96% of EdTech platforms have been found sharing student data with third parties. Igris protects student records and research IP at the prompt layer, isolates each lab and district, keeps spend inside grant and district allocations, and generates FERPA, COPPA, and GDPR documentation.
AI Tutoring Platform
An AI tutoring platform deployed in K-12 classrooms handles student data that FERPA defines as education records: names, grades, assessment scores, IEP and 504 plan references, behavioral notes, and learning progress data. If any of this reaches an external LLM provider without redaction controls, the platform has violated FERPA's unauthorized disclosure prohibition — an enforcement obligation the Department of Education intensified in 2025 with mandatory proactive compliance certifications. COPPA's revised 2025 rules, requiring full compliance by April 22, 2026, add explicit protections for students under 13: formal written security programs tailored to the sensitivity of student data collected, explicit opt-in consent before any data sharing, and biometric data protections for adaptive learning features that track behavioral patterns. For EU students, GDPR Article 9 applies to any learning disability or health-related data in the tutoring context, and for Indian students, DPDP Act 2023 obligations apply. The operational risk is separate: a classroom of thirty students submitting tutoring questions simultaneously creates a concurrent load spike that, without rate limiting, can exhaust a district's monthly model allocation in a single class period.
Guard PII redaction strips student identifiers — names, student ID numbers, grade references, and learning disability designations — from every tutoring prompt before it reaches the LLM provider, protecting FERPA education records and removing the personal data categories COPPA's revised 2026 rules require documented security programs for. Model restrictions on each district's connection keep routine tutoring interactions on economical generation models, reserving premium models for complex adaptive learning tasks. Igris Lens tracks cost per school district connection, giving the platform the per-district spend data that accurate billing and grant compliance reporting require. Rate limiting on each classroom connection prevents concurrent load spikes from overwhelming the service or exhausting a district's allocation in a single session. For EU and Indian students, GDPR Article 9 and DPDP Act 2023 obligations are addressed by the same PII redaction layer.
Student names, grades, and learning disability information are protected before they reach any external provider — not by platform policy, by technical enforcement on every tutoring call. FERPA compliance is supported by the redaction and audit trail. COPPA's revised 2026 requirements are met by documented data minimisation controls at the prompt layer. Concurrent classroom loads are bounded by rate limiting. Per-district cost data is available for billing without manual provider invoice reconciliation.
AI Research Assistant
University research teams using LLMs for literature synthesis, data analysis, and paper drafting work with material that carries both intellectual property and grant compliance obligations simultaneously. Unpublished findings, pre-publication datasets, proprietary methodologies, and grant-funded discoveries represent months or years of funded research — if any of this is transmitted to an external LLM provider, it may be processed or retained in ways that compromise publication priority, the institution's IP position, or the data handling protocols that NIH, NSF, and DoD grant conditions increasingly specify for AI tool usage. Without connection isolation between labs, a prompt from Lab A referencing an unpublished compound or pre-publication dataset could appear in Lab B's context window during a different session. Budget visibility is a grant compliance requirement: AI spend must be attributable to the specific grant it was charged against.
Igris creates an isolated connection for each research lab, so no context, prompt history, or data from one lab's sessions can appear in another's — isolation enforced at the gateway layer. Guard custom content patterns catch internal project identifiers, unpublished compound names, pre-publication data markers, and grant-specific codenames before any research prompt leaves the network. Igris Lens tracks cost per lab connection and per grant identifier, giving the research operations and finance teams the per-grant attribution data that grant compliance reporting requires. Budget caps on each lab connection prevent any research team from exceeding its grant allocation, with Lens surfacing a usage alert before the cap is reached.
Unpublished research, pre-publication data, and proprietary methodologies never reach an external LLM provider without a documented technical control in place. Each lab's work stays isolated from every other at the connection layer. Grant-attributed AI spend is available per connection without manual cost allocation. When a grant administrator asks what AI tools were used in a funded project and what they cost, the Lens report provides the complete per-grant answer.
AI Student Assessment and Learning Analytics
EdTech platforms using AI to generate adaptive assessments, score open-ended responses, and analyse learning outcome trends process data at the strictest intersection of education privacy law. Individual student assessment scores, response patterns, and adaptive difficulty profiles are FERPA-protected education records. For students under 13, COPPA's revised 2026 rules extend protections to the behavioral patterns generated by adaptive systems. For EU students, AI systems that assess educational performance are classified as high-risk under the EU AI Act, with documentation, human oversight, and audit trail requirements from August 2026. If student assessment data reaches an external LLM provider without documented redaction controls, the platform has created a FERPA unauthorized disclosure event on every scoring call — potentially for an entire cohort simultaneously.
Guard PII redaction strips student identifiers from every assessment and analytics prompt, leaving the model the pedagogical content and performance pattern it needs without the identifying linkage that creates FERPA and COPPA exposure. The EU AI Act high-risk system audit trail is generated automatically for every assessment AI interaction — the technical documentation that EdTech platforms selling into EU schools must produce from August 2026. Rate limiting on assessment connections prevents concurrent batch scoring jobs from overwhelming the service during end-of-term assessment periods. Per-district cost tracking feeds accurate billing. GDPR and DPDP Act obligations for EU and Indian student performance data are addressed by the same redaction layer.
Student assessment data is protected by technical enforcement on every scoring and analytics call. FERPA compliance and COPPA 2026 security program requirements are supported by the redaction and audit layer. EU AI Act high-risk documentation exists from the first assessment the AI scores. Concurrent end-of-term assessment loads are bounded before they become service disruptions.
AI Grant Writing and Research Proposal Assistant
Faculty using LLMs to assist with grant proposal writing share some of the most competitively sensitive material in academic research: unpublished research plans, preliminary experimental data, proposed methodologies representing months of intellectual development, and funding strategies whose disclosure could affect priority claims in competitive grant cycles. If this material reaches an external LLM provider and is processed or retained, the institution's position in the funding competition — and in any subsequent publication priority dispute — is potentially compromised before the proposal is submitted. For multi-investigator grants involving collaborating institutions, the isolation requirement is structural: University A's unpublished research plan cannot appear in University B's proposal session.
Guard custom detection patterns catch internal project identifiers, preliminary data references, and unpublished methodology markers specific to the institution's research taxonomy, blocking them before any proposal prompt leaves the network. Per-investigator connection isolation ensures that one faculty member's proposal context is fully separated from every other's — no cross-investigator or cross-institution context contamination. Budget caps per grant connection attribute AI spend on proposal writing to the correct funding source. The audit trail records every proposal development interaction, providing the institutional IP record the technology transfer office needs if a priority dispute arises.
Grant proposal content stays inside the institution's governance perimeter. Each investigator's work is isolated at the connection level. Proposal development AI spend is attributed to the correct grant. When a priority dispute arises, the governance log provides a timestamped institutional record of when the research material was processed — and through which governed, isolated channel.
Govern AI without exposing students or research
See how education and research teams redact FERPA data and unpublished IP, isolate labs and districts, and hold spend to grant budgets.