What we do
Proof, not promises.
Most AI goes live tested like ordinary software, without limits, evidence or an owner. We bring the assurance that regulated and safety-critical settings demand.
Our method
Four steps to trusted AI
Step 1
Test
We test AI systems against the hazards, laws and standards that apply to them, and attack their guardrails before anyone else can.
Step 2
Guard
We design hard limits outside the model, clear guidance inside it, and safe fallbacks for when things go wrong.
Step 3
Certify
We state in writing the trust level a system has earned, for what use, in what envelope and on what conditions.
Step 4
Monitor
We keep watching as models drift and rules change, and re-test so that trust is kept, not just won.
Trust levels
Autonomy is earned, one level at a time
We use six trust levels, from information only to full autonomy. A system moves up one level at a time, each with its own evidence, and moves down again if it drifts or changes without assessment.
| Level | Name | The AI system | The person responsible |
|---|---|---|---|
| T0 | Information only | Shows information and takes no action. | Decides and acts. |
| T1 | Advisory | Recommends actions. | Decides every action. |
| T2 | Supervised action | Acts when a person approves each action. | Approves each action. |
| T3 | Conditional autonomy | Acts alone inside a defined envelope. | Monitors, and takes over on request. |
| T4 | High autonomy | Acts alone and reaches a safe state on its own if things go wrong. | Sets goals and reviews. |
| T5 | Full autonomy | Acts alone in all conditions it claims to handle. | Provides oversight and accountability. |
For some decisions, such as those about a child's welfare or any action that could cause harm to people, we will not certify above the level at which a person makes the decision.
What you receive
The NeoFelis Trust Certificate
At the end of an assessment you receive a written certificate you can show to regulators, buyers and insurers. It states:
- the trust level your system has earned;
- the use it is certified for, and the operating envelope it must stay within;
- the conditions attached, including the named person in your organisation who is accountable for it;
- what was tested, and what was not: anything outside the envelope is outside the certificate;
- an expiry date, and the changes or incidents that trigger a re-test.
We do not build the systems we certify, so the evidence is independent. Where a sector has its own approval route, our certificate supports it rather than replacing it.
Expert services
Knowledge on demand
Alongside assurance, our team and Board of Advisers offer:
White papers and research publications
Thought leadership written with our advisers for boards, regulators and investors, followed by peer-reviewed research.
On-demand expert advice
Direct access to our advisers for clinical, industrial, financial and technical reviews, contracted through NeoFelis Labs.
Forward-deployed engineering and advisory
Engineers and advisers who work alongside your team to take AI from pilot to production, safely.
Research and domain knowledge for AI
Specialist knowledge and evaluation for organisations building AI in regulated fields.
Expert engagements are contracted through NeoFelis Labs.
How we work with you
Trust is earned, then kept
01
Listen
A trust conversation: what stops your most promising AI from reaching production?
02
Earn
A focused assurance pilot on one system, producing an evidence pack you keep.
03
Certify
A full assessment and the NeoFelis Trust Certificate at the level your system has earned.
04
Keep
Assurance as a service: monitoring, re-testing and renewal as your system and its world change.