EU AI Act
Start with the intended use.
Review the agent’s purpose, risk classification and relevant duties, including human oversight, transparency and records.
Agent Governance Benchmark
Benchmark your agents against the rules that apply.
Assess your agents against any regulation or policy—from the EU AI Act and regional requirements to industry standards and your own company rules. See the gaps, the evidence behind each finding and what to fix.
The right rules for the right agent
Its job. Its industry.
The places it operates.
Use any regulation, standard or company policy your business needs to assess. Apply it in the context of the agent’s job, industry and the places it operates, then turn the findings into practical next steps.
Regional regulations
Account for where it operates.
Bring the requirements of the countries and markets involved into the review, alongside the agent’s data and decisions.
Industry requirements
Reflect the work your sector does.
Assess the relevant sector rules, standards and oversight expectations against the task the agent is meant to perform.
Your company policies
Apply the rules your business sets.
Review decision rights, data access, financial limits, approval routes and internal policies in the same assessment.
From a requirement to an action.
See the rule behind a finding, the evidence available, what needs to change and who owns the next step.
The control benchmark
Know what you are checking.
Agents and owners
Which agent is running, who owns it and what the assessment covers.
Decisions and limits
Define its Business Authority: what the agent may decide and do, its limits and when it must stop or ask a person.
Data access
The information it needs and the records it may change.
Human review
Who approves exceptions and how someone can intervene.
Records of the work
What was requested, what was approved and what the system recorded.
Testing and updates
Tests for the current version and any proposed changes.
What the benchmark delivers
From a finding to a practical fix.
Bring your agents into view
See the agents, versions and owners available through your connected platforms. Missing information stays visible.
Understand what is missing
Review the rules, permissions, approvals and records behind each agent. See what is confirmed and what still needs checking.
Decide what to fix
Agree which fixes matter most, who owns them and how to test them. Update the plan as agents change.
Clear findings
See why a finding was raised.
| Benchmark findings | What it tells you |
|---|---|
| Material exposure | Evidence supports a significant control gap. Review the affected action before expanding its responsibility. |
| Remediation needed | A specific gap is supported within the assessed scope. The recommendation explains what to change and how to validate it. |
| Controls supported | The checked controls are supported for the stated scope, version and evidence window. |
| More evidence needed | Access, context or behavioural proof is incomplete. The uncertainty stays visible. |
Company policy · Purchasing agent
A spending rule that needs a reliable approval step.
- The requirement
- Orders above the company’s spending limit need the purchasing owner’s approval.
- The finding
- No approval threshold is visible in the agent configuration. Check whether it is enforced elsewhere.
- The next action
- Confirm the control with the owner. Where it is missing, add the approval route and retest an order above the limit.
Keep the benchmark current.
Daily refresh
See new agents, changed versions and updated findings, with the changes since the last assessment.
Reports you can use
Share the agent list, findings and supporting records with the people who need to act.
Recommendations you can act on
Get specific recommendations for the agreed scope. Then decide who will carry out the work.
The output
See an example of the report.
An overview shows where to focus. Open a finding to see its evidence and the next decision.
ILLUSTRATIVE BENCHMARK OUTPUT
Agent overview
- Agent
- Purchasing agent · sample
- What it covers
- Connected runtime and available configuration
- Owner
- Purchasing team
- Review focus
- EU AI Act, applicable policies, authority and evidence
ILLUSTRATIVE BENCHMARK OUTPUT
Finding → next decision
- Observed
- No approval threshold visible in configuration
- Evidence
- Configuration review; enforcement not yet tested
- Next step
- Confirm existing controls with the owner
- Validate
- Test an order above the agreed limit
Decide how to put the fixes in place.
A clear handover to the people who make the changes.
The benchmark gives your team the findings and improvement plan. Your team or a delivery partner implements the changes. With the Novagentica platform, they can test, review and redeploy the approved version, then check its actions and recorded results.
Explore the full platform ↗Start with the agents you already have.
Tell us which platforms you use and what you need to understand. We’ll confirm what the assessment can cover.