The Curator is built by a team that believes AI should be useful, honest, and safe. This statement describes the principles and controls that shape everything we ship.
Our principles
- Human-in-the-loop. AI drafts; humans approve. Editorial content, moderation decisions, and account actions all involve human review.
- Transparency. Every AI-generated page is labeled. Reasoning, sources, and version history are visible to readers.
- Privacy by design. The minimum data required to deliver the experience, encrypted in transit and at rest, deletable on request.
- No dark patterns. Clear pricing, one-click unsubscribe, no coerced consent.
- Fair-use content. Images and quotations respect their source licenses; attribution is preserved.
Safety guardrails
- Rate limits on demo turns (per user and per IP) to prevent abuse.
- Content filters on generated output for hate, self-harm, and CSAM.
- Refusal to impersonate real people in first person, or fabricate quotes attributed to them.
- No medical, legal, or financial advice — educational framing only.
- Auditable moderation queue for user-submitted topics and corrections.
Privacy & user rights
- Data Subject Access Requests (DSAR / GDPR / CCPA) processed from your account page — approved requests receive a full ZIP export via email within 30 days.
- Account deletion removes personal data; content contributions are anonymized.
- Cookies: strictly necessary always; analytics and advertising cookies gated behind consent.
Incident response
Suspected security or safety issues can be reported via our contact page with the subject "Security". We acknowledge within 2 business days.
Related
The rules Curator operates under
What the assistant will not do
It will not invent quotations, attributions, publication dates, awards or works. A fabricated line attributed to a real writer is the worst failure available to a publication about culture, and the system is built to ask rather than guess.
It will not claim to have read or watched anything. It reasons from public writing and from what you describe, and it is designed to keep saying so rather than adopting the voice of someone who was there.
Taste, not manipulation
The assistant is not optimised for time-on-site. It is built to make a case and then accept a refusal, and 'you have probably already read this, so here is the less obvious one' is a designed behaviour rather than an accident.
It does not flatter. Agreeing with every stated preference would make it useless, so it is built to argue when the evidence in your own list contradicts what you say you like.
Accuracy and uncertainty
Where a reading is contested, competing views are presented with the evidence that would settle them. Confidence is stated in the text rather than implied by fluency.
Generated editorial passes a quality gate — sourcing, specificity, verified quotation, stated assumptions — and drafts that fail are held back. Material predating that gate has been pruned from the archive.
Data, privacy and training
Sessions and collections are private to your account, are not sold or published, and are not used to train third-party models. The data-request tab exports or erases everything we hold without a support ticket.
Anonymous aggregate counts keep the service running; nothing identifying a person leaves the system for that purpose.
Human accountability
A named publisher answers for everything here, including anything the pipeline produced. If it is on the site, a person approved it.
Reader corrections are the safety mechanism we actually rely on — easy to file from any essay, read by a human, and answered with either a change or an explanation of why not.