AI Risk
EN
Get in touch

AI Vanguard · testing service for AI agents

Targeted testing for AI agents. With a separate report.

We test what matters for your use case, on site if needed, check how agent and team work together and, where it fits, set up an automated test setup with monitoring. Once, with a report and an introduction.

Test report · AI Vanguard

Agent: customer support · findings

  1. Tool permissions24 cases tested · no deviation No finding
  2. Approval in the team2 approvals without review Note
  3. Data leaking between customers1 case above threshold · fix before continued operation Urgent
Example with made-up data.

Service

What AI Vanguard does for you

Targeted testing

We test what can cause harm: permissions, data leakage, prompt injection, behaviour in sensitive cases. Remotely, or on site when needed.

Checks at team level

Not just the agent: who approves, who reacts to a fault, who can stop the agent? We check how it works together with your team.

Automated setup

Where it fits, we install and build an automated test setup at your company and set up monitoring and alerts following best practices.

Separate report

A separate report: what was tested, how, with what result and what was not tested.

How it works

How an engagement runs

  1. Call and scope

    You describe the agent and its use. In a meeting we suggest a package and agree the scope together.

  2. Test plan

    Which risks, which cases, which thresholds. Your company approves the plan.

  3. Test and set up

    Targeted checks, remotely or on site, the team-level review and, where it fits, the automated setup with monitoring.

  4. Report and handover

    A separate report with findings and recommendations, and an introduction for your team if you wish. Then the engagement is complete.

Packages

Three packages for businesses

Each package is a single engagement. The minimum duration shows the smallest scope; we clarify the exact effort in a meeting. No subscription, no ongoing services.

Testing days

From 2 days

Targeted checks on your agent.

  • Checks on the risks we agree on
  • Remote, or on site if needed
  • At least 2 days
Book a meeting

Assessment

From 1 week

Test protocol with a detailed plan and explained results.

  • Detailed test plan for the main risks
  • Test protocol for every step
  • Results explained, with traffic light
  • At least 1 week
All details
  • What was tested, how and with what result
  • What was not tested
  • Recommendations for next steps
  • Introduction for your team if you wish
Book a meeting

Most complete

Complete package

Test trail as code

From 3 weeks

Monitoring, guardrails and consulting included, tested and documented.

  • Everything in the assessment
  • Monitoring and alerts set up
  • Guardrails for the agent set up and tested
  • Consulting for your team
  • Every step documented
  • At least 3 weeks
All details
  • Setup installed and built at your site, following best practices; your company runs it
  • Every test step versioned as code and traceable
  • Findings as a chart in the report
  • Green: no deviation in the tested cases
  • Yellow: a deviation your company should look at
  • Red: a deviation above the threshold, fix before continued operation
  • Guardrails reduce risks but do not rule them out
  • After handover and introduction the engagement is complete
Book a meeting

Clear boundaries

What AI Vanguard is and is not

AI Vanguard is an additional testing layer next to your own tests and controls. What is tested is what the test plan says. Release and responsibility stay with your company.

In detail

What AI Vanguard does

  • Tests AI agents against a test plan
  • Checks how agent and team work together
  • Where it fits, sets up an automated setup with monitoring
  • Records findings in a separate report and hands it over

What AI Vanguard is not

  • Not a validation, sign-off or certification
  • Not proof that a system is safe or works correctly
  • Not legal advice
  • No substitute for your own testing, approvals and responsibility
  • No ongoing service: after report and introduction, the engagement is complete

Frequently asked questions

What is AI Vanguard?

AI Vanguard is a testing service for AI agents. We test against a test plan, on site if needed, check how the agent works with the team and deliver a separate report. It is an additional testing layer, not a validation or a sign-off: what the report does not mention can still be faulty.

Is this a subscription?

No. Each package is a single engagement. After handover and introduction it is complete, with no ongoing services.

How do we start?

With a meeting. In it we clarify the agent, its use and the benefit for your company, and suggest a package: testing days from 2 days, assessment from 1 week, complete package from 3 weeks. We only start once you have confirmed the engagement.

What is the automated setup?

Where it fits your use case, we install and build an automated test setup at your company and set up monitoring and alerts following best practices. After that, your company runs the setup itself; we hand it over with an introduction. Whether it fits is decided in the test plan.

Do you come on site?

Yes, if systems, data or processes require it. Otherwise testing runs remotely. We are based in Bözberg near Brugg, canton of Aargau, Switzerland.

What does a team-level review mean?

We test not only the agent but also the processes around it: who approves, who reacts to a fault and how the agent is stopped. What is tested is what the test plan says.

What does red mean in the report?

Red means a deviation above the threshold in the test plan was found in the tested cases. The report recommends fixing it before continued operation. Whether the agent is stopped or changed is decided by your company.

Does AI Vanguard replace our own tests?

No. AI Vanguard adds to your own tests, approvals and controls. What is in the test plan is tested; what is not, is not tested.

Does AI Vanguard help with the EU AI Act?

Your company can use the report and the documented test steps for its own documentation. Whether a system meets legal obligations is for your company to assess with its legal advisers. AI Vanguard is not legal advice and not a certification.

What happens to our data during testing?

The details are set out in the engagement. The principles are in the privacy policy.

Let’s set up a meeting

Describe your agent in two sentences: what it does and which tools it uses. We get back to you with a first assessment and a suggested date.

info@main-skill.ch
+41 76 511 52 25