THREE THINGS DAWN DOES NOT DO
Most of what people fear about institutional AI is what Dawn refuses to do at the architecture level. These are not commitments in a policy document. They are constraints in the code itself.
Dawn does not predict individuals.
No person-level prediction layer. No model that scores you, your family, or anyone you know. The system operates at the level of patterns, places, periods and institutions. Never at the level of named human beings.
Dawn does not score people.
No risk score. No flag. No tier. No rank. Institutions use Dawn to understand the context they are working in. They do not use it to label the citizens they serve.
Dawn cannot be turned into a surveillance tool.
The request-gating layer rejects any query shaped like individual lookup, individual ranking, or individual tracking. Same code path for everyone, including us. Not toggleable from an admin panel. Not negotiable.
Every output is advisory. In plain English. Aimed at the institution, never at the citizen.
Available on request.
Anything procurement might need, ask. We would rather write you a one-page note than maintain a hundred-page PDF nobody reads.
DPIA outline. Architecture diagram. Data Processing Agreement (DPA). Sub-processor list. Vulnerability disclosure policy. Vendor security questionnaire (SIG, CAIQ, or your own template). Penetration-test scope statement. Hosting, security and data-processing detail.
Dawn Intelligence Ltd.
Registered with Companies House (England and Wales).
Registered office, 3rd Floor, 86–90 Paul Street, London EC2A 4NE.
SEIS / EIS Advance Assurance issued by HMRC, May 2026.
