Five dimensions. One framework.
“Is the AI safe?” is not one question. We evaluate and build AI systems against five distinct dimensions — Secure, Private, Reliable, Accountable and Accessible — each with its own service, its own tests, and its own named risks.
One question each, answered differently.
Four map directly onto a group in the periodic table of AI risks. The fifth — Accessible AI — doesn't, because it's not a risk to mitigate so much as a requirement to meet.
Secure AI — Can it withstand an attacker?
From prompt injection to model theft and runaway agents — the attack surface unique to AI systems, not just the network around them.
Private AI — Does it respect the data it touches?
What it was trained on, what it retrieves, what it remembers — and whether anyone gave permission for any of it.
Reliable AI — Does it do what it says, consistently?
Hallucination, drift, sycophancy, brittleness to rewording — the gap between a demo that worked once and a system you can depend on.
Accountable AI — Who owns it when it's wrong?
A named owner, an audit trail, a classification against the EU AI Act — the governance layer that turns “the AI did it” into an answerable decision.
Accessible AI — Can everyone actually use it?
This dimension sits outside the risk table by design — it is about what a system must affirmatively do, not only what it must avoid.
Underneath: the USER PASS engineering pillars.
These five dimensions are about AI trust — whether a system should be deployed and how it's governed once it is. USER PASS is something else: the 8-pillar engineering-quality framework VVnT Converge runs during delivery and VVnT Accreditation independently certifies against — the plumbing underneath, not a sixth dimension here.
Five dimensions. Five places to go next.
A few things people ask us.
What is the AI Trust Framework?
VVnT SeQuor's framework for evaluating and building AI systems across five dimensions: Secure, Private, Reliable, Accountable and Accessible AI, each tied to a specific service and a specific set of named risks.
How does this relate to the periodic table of AI risks?
Four of the five dimensions — Secure, Private, Reliable and Accountable — map directly onto groups in the periodic table of AI risks. Accessible AI sits outside that risk table by design: it is about what a system must affirmatively do, not only what it must avoid.
Is this a certification?
No. VVnT Accreditation is our independent assurance service that can assess a system against these dimensions and issue a VVnT POC Assurance Certificate. The framework itself is a way of organizing that work, not a credential on its own.
Know where your AI stands across all five.
Talk to VVnT SeQuor about a walkthrough against this framework — or start with the free V-SeQuor assessment if you're a small team.