Trust
Trust and procurement
For the IT, privacy and procurement people reviewing Academia Superiora. Everything here describes the service as it runs today. Where something is planned and not yet done, we say so.
At a glance
| Area | Today |
|---|---|
| Hosting | Amazon Web Services, us-east-1 (N. Virginia, United States) |
| Certifications | None yet: no SOC 2 report and no ISO 27001 certificate. Our HECVAT 4 answers are available on request. |
| Encryption in transit | HTTPS (TLS 1.2 or 1.3) between your browser and the service, and to every provider we use |
| Backups | Database copies in Amazon S3 are encrypted (SSE-S3). Encrypting the server’s own disk and its daily snapshots is planned and not yet done. |
| Passwords | Stored only as salted PBKDF2-SHA256 hashes (390,000 iterations) |
| Sessions | A random token in an HttpOnly cookie, kept 24 hours; we store only its hash. The app never keeps a copy in browser storage. You can see and end your sessions in Settings. |
| Sign-in | Email and password, with two-step sign-in by an authenticator app, which an institution can require. Sign in with Google or Microsoft is planned. |
| Single sign-on (SAML) | SAML 2.0 per institution (Okta, Entra ID, Shibboleth/InCommon and others), through WorkOS, turned on for an institution when it asks. An institution can require it; administrators it names keep a password sign-in for emergencies. |
| Taking your data out | An institution administrator can download the institution’s shared projects and its own records as one ZIP in Settings at any time: JSON and CSV for each kind of record plus the original files, with a README. A member’s private projects are not in it. When an institution leaves, we hand over all of its content within the 30 days before deletion. Every download is recorded in the audit log, and secrets such as password hashes are never included. |
| AI database access | Queries the AI writes run under a read-only database role that can read only public market data. The database enforces the limit, not the prompt. |
More detail is on the Security page.
Service hours
Weekdays, 12:00 to 04:00 UTC (the next morning): 06:00 to 22:00 US Central in winter, 07:00 to 23:00 in summer.
Outside those hours, and from Saturday 04:00 to Monday 12:00 UTC, the service is switched off: nothing is lost and nothing is running. The hours are fixed in UTC, so the US Central times move by one hour when the clocks change.
We do not publish an uptime figure. A service level is offered only in a signed Order Form.
Where your data is
Where it rests
- One PostgreSQL database on our server in AWS us-east-1 holds accounts, projects, chats, meetings and transcripts, votes, the audit log and the public warehouse. Uploaded files are stored inside that database; there is no separate file store.
- Recorded decisions are also copied into a graph database (Neo4j) on the same server.
- Copies: encrypted weekday database dumps in Amazon S3 and daily snapshots of the server, both in us-east-1.
Where it goes
- To our AI providers, for the request you made (see “AI and your data”).
- To Recall.ai, for a meeting you ask the service to listen to.
- To Google or Microsoft, when you connect a calendar or sign in with them.
- To WorkOS, when your institution signs in with single sign-on.
- To Amazon SES, for email the service sends you; to Sentry, for error reports with personal data collection off.
Subprocessors
The subprocessors page says what each one does, what data it sees, where and when. Today’s list:
- Amazon Web Services, Inc.
- Amazon Web Services, Inc. (Amazon SES)
- Anthropic, PBC
- OpenRouter, Inc.
- OpenAI models, through OpenRouter
- TypeSafe models, through OpenRouter
- Google LLC (Sign in with Google)
- Microsoft Corporation (Sign in with Microsoft)
- WorkOS, Inc.
- Recall.ai, Inc.
- Google LLC (Google Calendar)
- Microsoft Corporation (Microsoft 365 / Outlook)
- Tavily
- DuckDuckGo
- Functional Software, Inc. (Sentry)
- Google LLC (Google Analytics)
- Stripe, Inc.
Two things are not on this list on purpose. Anaplan, for institutions that connect it, is the institution’s own vendor: we read the models it chooses, with its own credentials. Product-usage events are kept in our own database; if we turn on product analytics, it runs on a server of ours, not a third party’s.
AI and your data
- Athena, the assistant in Academia Superiora, writes with Anthropic’s Claude models. We call Anthropic directly; if that call fails, we reach the same models through OpenRouter.
- A few smaller jobs use other models through OpenRouter: search embeddings and plain-language institution search (OpenAI models) and sorting the decisions found in a meeting (TypeSafe). The subprocessor list names each one and what it receives.
- Every request we send through OpenRouter asks it to use only hosts that do not keep or train on the data.
- We do not use your content to train AI models. Anthropic’s Commercial Terms do not allow Anthropic to train on it.
- We send a model only what the request needs: your question, the parts of the project the answer depends on, figures from the public warehouse and, in a project chat, that project’s files.
- Files sent to Anthropic to build a document or a deck are deleted from Anthropic’s file storage when the build ends.
- AI output is a draft for a person to review. Nothing the AI finds in a meeting is written to a project until a person approves it, and votes are cast by people.
FERPA
Academia Superiora is built for institution-level planning. Its warehouse holds public, aggregate federal and state data (IPEDS, College Scorecard and other public sources), not records about individual students.
We do not need student education records to do our work, and our Terms do not allow uploading them without a written agreement with us. If an institution needs to share them, that agreement makes us a “school official” with a legitimate educational interest under FERPA, under the institution’s direct control, and the Data Processing Addendum applies.
Users can upload files and record meetings, so an institution decides what enters the service. A transcript can name a person. We process that content only to provide the service, on the institution’s instructions.
Each institution’s projects, files, meetings and votes are visible only to its members and to the people a project is shared with.
How long we keep data
| Data | Kept |
|---|---|
| Your projects, files, chats, meetings and votes | While your account is open. A deleted project, file or meeting can be restored for 30 days, then it is permanently deleted. |
| Product-usage events (which pages and features are used) | 90 days, then deleted automatically |
| Audit log (sign-ins, admin changes, access tokens) | two years, then deleted automatically. Until then a record cannot be edited or deleted: the database refuses. |
| AI connector call log (IDs only, never the question or the answer) | 400 days, or as long as the project it names is active |
| Meeting recordings at Recall.ai | 7 days, then deleted by Recall.ai. The transcript is kept in your project. |
| Files sent to our AI provider to build a document or deck | Deleted from the provider’s file storage when the build ends |
| Database backups | Weekday copies kept 30 days in Amazon S3; daily server snapshots kept 7 days |
| AI usage records (model, tokens and cost; never the text) | Kept as billing records |
The Privacy Policy describes deletion, including deleting an account.
Incidents
If we confirm a security incident that affects your institution’s data, we tell your institution’s contacts without undue delay after confirming it: what happened, what data it touched, what we have done and what you may need to do.
To report a vulnerability, write to hello@superiora.ai. Our security.txt and the Security page say how.
Documents on request
- A Data Processing Addendum, with a Public Institution Addendum and the subprocessor annex
- Our HECVAT 4 answers
- Our incident-response and breach-notice policy
- An architecture and data-flow description
Write to hello@superiora.ai.
If anything here is unclear, write to hello@superiora.ai.
