01
Cited reasoning, every output
Every risk score, every migration playbook, every runtime block carries the source evidence with it. We do not ship un-cited model output.
Trust Center · Responsible AI
Applied to our own AI first.
Quanterios sells AI governance. We have to live the discipline ourselves. The principles below shape how the AI Decision Engine is built, deployed, and audited.
01 · Six principles
02
Deterministic where it matters
Risk scoring is rule-based + XGBoost, auditable, reproducible, never hallucinated. LLM reasoning sits on top, not underneath.
03
Customer data is not training data
We do not fine-tune our models on customer content. Our migration-outcomes corpus uses anonymised aggregate signals only.
04
Human-in-the-loop for irreversible actions
Crypto Agility API algorithm swaps require explicit policy authorisation. AI Runtime denials are auditable and customer-overridable.
05
EU AI Act discipline applied to ourselves
We classified the platform's own AI components against the EU AI Act risk tiers and meet the obligations for the tier we sit in.
06
Pre-trained model provenance
Third-party LLMs (Claude, OpenAI fallback) with fully signed contracts, EU residency where available, no-training agreements. Versions pinned in our internal AIBOM.
02 · Internal classification
Quanterios's own AI components, what they do, the risk-tier classification we hold ourselves to, and the obligations we meet.
AI Decision Engine · scoring
Limited-risk · transparency · human-oversight gate on irreversible recommendations.
AI Decision Engine · briefings
Limited-risk · transparency · cited evidence chain · customer-overridable.
AI Auditor · code scanning
Limited-risk · transparency · suggested fix is advisory, not enforced.
AI Runtime classifier ensemble
Operational · auditable · denial-with-evidence to SOC · customer-overridable policy.
Need the responsible-AI package?
Email trust@quanterios.com for the technical documentation, classification rationale, and post-market monitoring summary.