When is an assessment useful?
- You are selecting or changing a model for a real workload.
- You want evidence before deploying a customer-facing or internal AI system.
- Your agent uses tools, handles sensitive information or needs to respect approval limits.
- You need to understand a recurring failure, recovery behaviour or gap in evaluation coverage.
- You are a model provider seeking controlled, scenario-specific evidence.
What can be assessed?
| Area | Questions we examine |
|---|---|
| Workflow capability | Does the system complete the agreed task against independently checked criteria? |
| Reliability | What happens with edge cases, repeated attempts and controlled faults? |
| Security | How does it respond to scoped injection, data exposure, unsafe tool use and policy-bypass tests? |
| Organisational rules | Does it respect escalation, approval limits, allowed sources, tool authority and data handling? |
| Operational behaviour | What do valid latency, availability, token usage and cost observations establish? |
| Governance evidence | Which technical findings can support your chosen internal review process? |
What do you receive?
- An agreed workflow, requirements and scenario scope.
- Declared rules, critical failures and acceptance checks.
- Controlled execution records and sanitised evidence where applicable.
- Eligible pass/fail outcomes separated from invalid or unavailable measurements.
- Findings, explicit exclusions, coverage gaps and recommended next validation steps.
- A report for technical, product, risk and governance stakeholders.
The exact deliverables, fees and timing depend on the agreed scope. Customer scenarios and artifacts are not published without written approval.
What should your request describe?
- Workflow: the people using it, the job to complete and the outcome you need.
- Target: the model, application or agent and its relevant configuration.
- Rules: permissions, approved sources, escalation and actions that are never allowed.
- Risks: the failures that matter most in practice.
- Data and access: synthetic fixtures, documentation or sandbox access available for authorised testing.
- Success criteria: the checks and thresholds relevant to your decision.
- Stakeholders: the technical, operations, legal, privacy, security and domain reviewers.
- Timing and constraints: your window, budget context, procurement or confidentiality needs.
How does the request flow work?
- Complete the request form on this website. Unknown details can be refined during scoping.
- Review the summary and submit it privately to the VIS assessment team.
- A receipt appears here after the email service accepts the request.
- We review fit and arrange an appropriate private channel to agree the engagement.
Share a high-level brief. Do not include passwords, API keys, personal records or proprietary documents. Submitting a request does not authorise testing or create an engagement.
Good questions.
What is an AI Assurance assessment?
A scoped technical evaluation of an AI system against a declared workflow, configuration, rules and success criteria. The findings show observed strengths, failures, evidence gaps and limitations.
Does a request authorise testing or create a contract?
No. We agree scope, authorised data and access, confidentiality, timing and commercial terms before testing starts.
Does an assessment certify compliance?
No. Technical findings support review by your governance, privacy, legal, security and domain owners. They do not certify compliance or guarantee safety.