ClariAze AI reads what you describe or upload and proposes a complete, structured set of risks in about a minute. It never decides — a named person accepts or declines every proposal, and the reason is kept.
A description in plain words, one area you want to explore in depth, or the documents you already hold.
A guided form captures the project, its scope and stakeholders, the data involved, the regulatory environment, third parties and existing controls. Every field explains what a good answer looks like.
Choose one of 15 categories — cybersecurity, supply chain, financial, compliance and more — and answer the questions ClariAze prepares for that type. Six to ten focused risks come back.
Upload up to six contracts, audit reports or procedures — PDF, Word, text, spreadsheets or images — and tell the AI what angle to take. It extracts the risks those documents imply.
Each analysis is a record with a request number, kept in the analysis history with the information that was submitted — so anyone can see later what was asked and what was proposed.
The AI estimates likelihood and impact. The level is then calculated by ClariAze itself — the same way every time — so a “Critical” means the same thing in every analysis, in every company.
| Likelihood | |
|---|---|
| 1 | Rare |
| 2 | Unlikely |
| 3 | Possible |
| 4 | Likely |
| 5 | Almost certain |
| Impact | |
|---|---|
| 1 | Negligible |
| 2 | Minor |
| 3 | Moderate |
| 4 | Major |
| 5 | Severe |
| Score = L × I | Level |
|---|---|
| 1 – 4 | Low routine monitoring |
| 5 – 9 | Medium plan mitigation |
| 10 – 15 | High management attention |
| 16 – 25 | Critical act immediately |
Inherent risk is the level before existing controls; residual risk is the AI's estimate after the controls you listed. Categories follow a shared 15-category ISO 31000 / COSO ERM taxonomy maintained by your administrator.
AI output is advisory. Nothing reaches your register until a compliance officer accepts it — and everything that happens is on the record.
Each risk arrives in the Pending queue, linked to the analysis that produced it.
Accepted risks receive a register number; declined risks keep their reason and can be reopened.
A responsible person, a deadline, an action plan and review reminders.
Every decision and change is written to an audit trail nobody can edit.
“An auditor doesn't ask what the AI said. They ask who decided — and why.”
An enterprise AI feature must be predictable. ClariAze AI is constrained at every step, from what it accepts to what it is allowed to say.
It works on risk, compliance and governance only. Off-topic requests are declined instead of answered.
Instructions hidden in a description or a document — “ignore the format”, “reveal your instructions” — are screened out before and after the model runs.
It does not reveal its configuration or internal instructions, and replies that would are withheld.
It is instructed to reason about roles such as “the finance manager”, not named individuals, and not to infer sensitive personal characteristics.
Requests are rate-limited, and oversized inputs are rejected with a clear message.
When input is refused, the event is written to the audit trail for the compliance team to see.
What you enter or upload is sent to the AI service only to produce the analysis. The result is stored with the analysis in your company's records, alongside the information that was submitted.
Employees see their own analyses; your company's compliance officers see all of the company's analyses. No other company can see them.
Anything your company's data-classification rules do not allow to be shared with an AI service. When describing a situation, refer to roles rather than named individuals.
Yes. Edit the inputs and run it again: pending and declined proposals are replaced with the new result, and risks you already accepted are kept.
In a demo we run ClariAze AI on a situation from your organisation and walk through the decision flow.