Aller au contenu principal

TRUST · GLASS-BOX AI

IA boîte de verre.Vérifiez par vous-même.

Chaque décision IA sur SENTINEL est chaînée par hachage dans un registre en ajout seul. Cette page permet à quiconque — client, auditeur, sceptique — de recalculer la cryptographie en direct contre notre base de production. Sans compte. Sans confiance requise.

— 01 / VERIFY A DECISION —

Don't trust us.Check the math.

Paste any decision ID from your SENTINEL workspace. Our production database recomputes the SHA-256 chain hash in front of you and tells you whether the record is intact and correctly linked to its predecessor. Only hashes leave the database — never your data.

No ID at hand? — the first signed record, written the day this page shipped.

— 02 / HOW IT'S BUILT —

Safety isn't a feature.It's the spine.

Hash-chained audit log

Every recorded action is hashed with SHA-256 over a canonical payload and chained to the previous entry. Rewriting history breaks the chain — visibly, for anyone who checks.

Append-only by construction

Database triggers reject UPDATE and DELETE on the audit log. Not a policy promise — a constraint the database itself enforces.

Humans approve high stakes

Outbound commitments — a bid, a client-facing message, a payout — queue for human approval before execution. The AI proposes; you dispose.

Budgets and speed limits

Per-minute, per-hour, per-day caps on every automated action, plus hard spend budgets per agent. A runaway loop hits a wall, not your wallet.

Tenant isolation

Row-level security on every multi-tenant table. Your data is invisible to other tenants at the database layer, not just the application layer.

Loi 25 & PIPEDA posture

Built in Québec under Loi 25: consent records, access and erasure workflows, data residency awareness. SOC 2 Type II is on the roadmap — we publish progress honestly.

— QUESTIONS ABOUT OUR POSTURE? WRITE TO HELLO@SENTINEL-OS.CA — WE ANSWER SECURITY QUESTIONNAIRES PERSONALLY —