“A vendor holds data on a hundred thousand of my students. I cannot see what they have, I cannot see who inside their company has looked at it, and when they get breached it is my name in the letter to families.”
The Security Model
Provider Vetting
The security model protects the data. Vetting governs who is allowed near it.
Commitments
We do not train models on student data.
We do not sell student data and we do not share it with advertisers.
We do not retain data a district asks us to delete.
Because districts bring their own key, the model runs in the district's own provider account under the district's own agreement. Anything learned in that district is a district trait and does not become a platform asset. AgenticED improves the platform, and it does not improve it by harvesting what it saw in one district and selling it to the next.
The audit log is append only by design. Audit metadata is retained for compliance while the underlying student data is deleted. Each entry is chained to the one before it, so any change to the log after the fact can be detected.
Compliance Posture
Independent third-party security audit in progress. The seven legal instruments, including the Data Processing Agreement and the Security Overview, are linked in the footer of every AgenticED product.
Subprocessors
The district brings its own model key, its own email, and its own SMS. AgenticED does not sit in the middle of a district's outbound messaging, so those sends never become a subprocessor relationship. The list is short by design.
| Subprocessor | Purpose | Data reached |
|---|---|---|
| Amazon Web Services | Compute (ECS Fargate, Lambda), database (RDS), storage (S3), key management (KMS), secrets, and transactional email | All platform data |
| Anthropic | Default model provider for agent reasoning | Prompt context. No PII where the JustData token architecture applies |
| Datadog | Observability for Board Response: metrics, traces, scrubbed logs | Identifiers, statuses, timings. Never message body, PII, or tokens |
| ElevenLabs | Text to speech, so agent replies can be read aloud | The text being spoken |
| Sign in with a Google account | Name, email address, and the sign in token | |
| HubSpot | Marketing and CRM: website forms and visitor analytics | Prospect contact data. No student data |
| Microsoft | Sign in with a Microsoft account | Name, email address, and the sign in token |
| OpenAI | Embeddings for knowledge base search (FAFSA, benefits, family resources) | Knowledge base content and the text of the question being searched |
| Bring your own key providers | OpenAI, Azure OpenAI, Google Vertex, Ollama, Groq, Together AI, used only when a district supplies its own key | Same as Anthropic, under the district's own agreement |
Co-browse is built by AgenticED and runs inside the AWS environment above, so it adds no subprocessor of its own. Last updated October 2026.
Responsible Disclosure
Report a vulnerability to support@agenticed.com. Every report is acknowledged.