Services / Packages
AI Assurance Test
Our most thorough test: an AI system run against 100 or more realistic scenarios, with a signed assurance letter.
From $5,000 per system2–4 weeks
Public agencies receive quotes structured to fit their procurement rules.
You own everything we produce.
Who this is for
Organizations usually call us about this when:
You’re launching a chatbot or AI assistant that customers, patients, or residents will use.
Leadership, a regulator, or a customer wants evidence the system was independently tested.
A vendor updated the model and you need to confirm it still performs.
Overview
Before people rely on an AI system, someone independent should test it the way real users will. We design 100 or more scenarios from your users’ point of view, such as a patient, member, customer, or resident, and test accuracy, safety, privacy, fairness, and when the system hands off to a person. You receive a full test report and a signed assurance letter stating what was tested, the criteria used, and the results as of the test date.
What’s included
- Test plan with 100 or more scenarios built around your real users
- Testing for accuracy, safety, privacy, fairness, and human handoff
- Documented results for every scenario
- Severity ratings and recommended fixes for each issue
- Retest of fixed issues before sign-off
- Signed assurance letter for leadership, regulators, or customers
Available under attorney-client privilege. We can perform this work at the direction of your counsel, so findings may be protected by privilege. Ask us about privileged engagements.
How it works
Plan
We agree on the system, its users, the risks that matter, and the pass criteria.
Test
We run every scenario, record the results, and rate any issues by severity.
Attest
After fixes are retested, we issue the test report and signed assurance letter.
How it works for your organization
The same service, tailored to the obligations and realities of each sector.
Private Sector
Companies use assurance testing to launch AI with evidence it works, which they can share with customers and partners.
- Customer-facing chatbots and AI assistants
- AI features inside products before release
- Evidence for enterprise customer due diligence
- Retesting after model or vendor changes
Public Sector
Agencies use assurance testing to confirm AI tools serve residents accurately and fairly before launch.
- Resident-facing chatbots and service tools
- Testing in plain language and multiple languages
- Accessibility and fairness across resident groups
- Documentation for councils, auditors, and the public
Nonprofits
Nonprofits use assurance testing to make sure AI tools treat the people they serve safely and respectfully.
- Client intake and support chatbots
- Testing with sensitive and crisis scenarios
- Protection of client information
- Evidence for funders and boards
Frequently asked questions
How much does AI Assurance Test cost?
AI Assurance Test starts at $5,000 per system. The final price depends on scope, and we confirm it in writing before any work begins.
How long does it take?
Most engagements take 2–4 weeks, depending on scope and how quickly information is available.
What do we receive?
The engagement includes: test plan with 100 or more scenarios built around your real users; testing for accuracy, safety, privacy, fairness, and human handoff; documented results for every scenario; severity ratings and recommended fixes for each issue.
Can we show that our system was independently tested?
Yes. Systems that pass testing can display a Cogent Assurance Tested seal, and you can publish a public summary stating what was tested, the criteria used, and the results as of the test date. Our methodology is published in the Resource Center.
Can this be done under attorney-client privilege?
Yes. When your counsel engages us and directs the work for the purpose of giving legal advice, our findings may be protected by privilege. Ask your counsel to contact us and we’ll structure the engagement with them.
What if a deliverable doesn’t meet the agreed scope?
We revise it until it does, at no extra cost. And if we miss an agreed delivery date, we take 10% off that deliverable.
Who owns the deliverables?
You do. Everything we produce for you is yours to keep, use, and share.
Related Guides
- AI Impact Assessment Example for Credit UnionsA worked example assessing an automated personal loan decision model, from purpose and data to risks, controls, and approval.
- AI Risk Register for Credit UnionsA sample register of common credit union AI uses, the data each one touches, its risk tier, the controls it needs, and who owns it.
- AI Tabletop Exercise Scenario for Credit UnionsA ready-to-run scenario where a member service chatbot exposes account details, with timed developments and questions for each phase of the response.
- AI Laws by StateWhere each state’s AI and privacy rules stand, with key dates and a page for every state.
Related services
AI Red Team Assessment
Adversarial testing that tries to make your AI system leak data, break its rules, or produce harmful output.
AI Agent Go-Live Review
A fast pre-launch check of an AI agent’s permissions, guardrails, and escalation, lighter than a full assurance test.
AI Risk Review
An independent review of one AI tool, vendor, or use case before you adopt it or expand it.
Talk through your AI and privacy questions
Tell us what you're working on, and we'll reply within one business day.