Inspect the evidence
Follow a memory or conclusion to its source. See what supports it, correct it, and keep uncertainty visible.
AI workflow assurance
Your AI gives answers. We help you check the evidence.
Testing the AI your business already uses. Guth assesses an existing AI workflow: its sources, permissions, and behavior in agreed test scenarios. You receive documented findings, supporting evidence, and prioritized next steps. Begin with one workflow, with scope and access agreed before work starts.
Assessments available for discussion · Product systems in development
Why it matters
Confidence needs something you can check.
Safety failures cost time, trust, and sometimes real money. Your customers may also need to understand how you manage AI. A scoped assessment gives teams a practical view of risk and a plan they can verify.
Set the permissions
Define tools, approvals, model choice, and where the agent must stop.
Keep the receipt
Record inputs, permissions, actions, and outcomes so failures can be investigated.
The assessment
A clear question, a controlled test, an honest finding.
We begin with the workflow, data, tools, people, and effects in scope. Pricing is quoted by scope after a short discovery conversation.
Agree on what to test
Map the use case, users, data, model, tools, approvals, expected behavior, and failure modes.
Test the workflow
Run agreed tests for unsupported answers, unsafe tool use, permission gaps, leakage, cost surprises, and recovery.
Report and repair
Deliver findings, evidence references, prioritized fixes, and checks you can run again after the fixes.
Regulation in context
Understand the rules that apply.
The EU AI Act uses a risk-based approach; the requirements that apply depend on your system, your role, and where it is used. NIST’s AI Risk Management Framework is voluntary guidance. We help you review your controls and records, while qualified legal advisers handle legal interpretation and formal compliance advice.
Names for the work
AI assurance, agent audits, evaluation, and governance.
You may be looking for an AI workflow assessment, an AI agent audit, AI risk assessment services, or help with agent governance. These requests overlap, but the question matters: what should the workflow do, who controls its actions, and what evidence would show that it meets your criteria?
AI evaluation tests behavior. AI governance defines owners, rules, and approvals. Observability supplies logs and traces. Assurance brings relevant evidence together to support a defined claim. A technical assessment does not automatically provide legal compliance, certification, comprehensive security testing, or ongoing monitoring.
Explore the AI assurance and discovery glossary for the terms people use, their differences, and the questions behind them.
Before you begin
Questions about an AI workflow assessment
How do you check AI answers?
Agree on representative examples, identify the approved sources and versions, and compare observed answers with the evidence and acceptance criteria. Record unsupported claims and uncertainty as findings.
Can my chatbot use old policies?
It can retrieve or repeat an obsolete source. Source provenance, retrieval quality, citation verification, and checks for stale knowledge help reveal that failure. The test scope determines which documents and cases are examined.
How do I test an AI agent before launch?
Define permitted and prohibited actions, run agreed scenarios, record the actual results, and retest after changes. An agent handoff checklist and customer acceptance tests can make the review easier to repeat.
How do I stop an agent sending without approval?
Enforce approval at the tool or action boundary, then test attempts to act without it. A written instruction alone does not demonstrate that the permission control works.
What will I receive?
The agreed assessment can include a workflow and source inventory, test criteria, observed results, evidence references, limitations, prioritized fixes, and a retest plan. Agree on the deliverables before access or work begins.
Does this include certification or continuous monitoring?
No blanket certification or legal compliance conclusion is included. Ongoing monitoring, broader governance programmes, security engagements, and additional workflows require their own scope. Start by describing the workflow and the question you need answered.