Loading
Every stance we've taken on security, privacy, data handling, and academic integrity, in one place — because if it isn't documented, it isn't real. Every page is technical, dated, and updated when the underlying decision changes.
From detection to affected-user notification. Written into the DPA, not just into a policy page.
Anthropic does not train on API traffic, and we fine-tune nothing on user content.
Parents and educators see effort, never the work. Enforced at the table rather than by a setting.
The claim worth checking is architectural, so here it is as a shape rather than as three paragraphs: what each stop holds, and what never reaches the next one.
A draft reaches local storage before it reaches us, so a refresh never costs you a paragraph.
Encrypted at rest, row-level isolated per account. Exportable and deletable from Settings at any time.
Not trained on. Limited retention for trust and safety under their standard terms — we hold no zero-retention agreement.
Never the attempt, never the tutor's reply, never the reflection. Enforced at the table, not by a setting.
A list of green ticks is the easiest thing on a trust page to write and the first thing a procurement reviewer checks. These carry their real state.
Every table carries a policy tied to the signed-in user, and a CI job proves one account cannot read another's rows.
TLS everywhere; Postgres and object storage encrypted at rest by the platform.
Dashboards read effort tables. The attempt, the reply and the reflection are not in them.
Written, versioned in the repository, and re-run when the architecture changes. Available to institutional prospects.
Nobody outside has tested us. When that changes, the summary appears here with its date and scope.
On the roadmap, not under way. We would rather say so than let “in progress” do work it has not earned.
Queued and undated. Today everything is stored in the US under Standard Contractual Clauses.
Trust is what happens when a company’s public promises match its private technical decisions. These six pages are the public promises; the code and infrastructure behind them are the private decisions. Both should match.
Infra, secrets, access, incident response — 72h disclosure window, live incident log.
What we collect. What we don't. What leaves. Zero training on your work.
Where your work lives, in what region, encrypted how, for how long.
Compatible with your school's honor code — by design, not by policy.
Effort signal, yes. Content, no. The enforcement is architectural.
Pilot workflow, DPA, class-level metrics, FERPA alignment.
The plain-english pages above are how we’d explain these commitments over coffee. The eight legal documents below are how our counsel wrote them for institutional review, procurement questionnaires, and DPO audits.
The legal agreement you enter when you use Brainback.
The legal version of /trust/privacy. Counsel-ready.
20-clause DPA for institutional and DPO review.
What Brainback may and may not be used for.
The three cookies we actually use.
Everyone who processes your data on our behalf. Dated.
COPPA-aligned under-13 workflow. Parental consent verification.
The technical detail behind /trust/security. For institutional review.
Every page in the trust center carries the date of its last substantive update. Every legal document carries a version number tied to a specific commit. When we make a material change — new subprocessor, new retention window, new security posture — we update the page and (for changes affecting existing users) email affected users. This is written into the DPA and it is a contractual commitment.
→trust
Six commitments · 8 legal docs · versioned + dated + updated.