Your AI system is technically ready. Is it humanly trustworthy?
The EU AI Act doesn't just require compliance documentation. It requires that AI systems can be understood, challenged, and overridden by the people who use them. That's a design problem — and most organisations aren't designing for it.
I developed the Trust by Design framework through research across 14+ AI implementation teams in the German public sector. It addresses exactly this gap.
What the Act requires from you
Most organisations are currently spending on legal and technical compliance. Almost none are investing in the human layer — the UX, explainability, and oversight design that determines whether compliance holds in practice. That is the gap this practice addresses.
The Problem
Technical compliance is necessary. It is not sufficient.
67% of users abandon AI systems they don't understand. 42% more helpdesk costs follow. And those are the consequences before an enforcement action.
The EU AI Act's human oversight requirements cannot be satisfied by a PDF policy or a "human in the loop" checkbox. They require that affected users can genuinely understand, challenge, and override AI decisions — and that the interface is designed to make that possible.
This is the Trust Layer: the designed human interface between your AI system and the people it affects. It is where technical compliance becomes real-world trustworthiness.
Explainability
Users can understand what the AI system did and why. Decisions are presented in plain language appropriate to the audience — not technical output passed directly to end users.
Recoverability
When the AI is wrong — and it will be wrong — users have a clear, designed path to recover. They can correct errors, reverse decisions, and escalate without hitting dead ends.
Accessibility
The system works for the full range of people it will actually be used by, including older adults, people with disabilities, and those with low digital literacy. Compliance and inclusion are the same requirement.
Human Control
Humans remain meaningfully in the loop. Override mechanisms are designed to be usable, not buried in settings. The AI supports human decision-making rather than replacing it by default.
What's Required
The obligations in plain language
The EU AI Act runs to over 400 pages. These are the obligations most likely to require design and research work — the ones your legal team alone cannot deliver.
High-risk AI classification
AI systems used in hiring, credit, healthcare triage, law enforcement, education, and public services are classified as high-risk. If your product touches any of these domains, the Act's full obligations apply to you from August 2026.
Human oversight requirements
High-risk AI must include meaningful human oversight mechanisms — not a disclaimer, but a designed capability for users to understand, challenge, and override AI decisions. This is a UX and design problem, not just a legal one.
Transparency and explainability
Users interacting with high-risk AI must be informed they are doing so and must be able to understand the basis of AI-driven decisions that affect them. Opaque outputs are non-compliant by default.
Fundamental rights impact assessment
Deployers of high-risk AI must conduct and document a fundamental rights impact assessment before deployment. This requires structured research with the populations affected — not a desk exercise.
Conformity assessment and CE marking
Many high-risk AI systems require a conformity assessment before they can be placed on the EU market. The documentation, testing, and technical file requirements are substantial and begin now.
Post-market monitoring
Providers must establish post-market monitoring systems that continuously track real-world performance, particularly for impacts on fundamental rights and safety. This must be designed in from the start.
How I Work
Three ways to work together
Trust Audit
I assess your AI system against the four Trust Layer principles and the specific requirements of the EU AI Act for your risk classification. You receive a documented gap analysis and a prioritised remediation plan your product and legal teams can act on.
Self-service tool at trustaudit.toolsFramework Implementation
I design and build the Trust Layer into your existing AI systems and processes — working alongside your product, UX, and compliance teams. The output is a system that is genuinely trustworthy, not one that passes a checklist on paper.
EU AI Act Readiness
I map your high-risk AI obligations, identify where your current system falls short, and build the human-centred compliance layer you need before August 2026. This includes documentation support, FRIA preparation, and human oversight design.
Who This Is For
Regulated industries deploying high-risk AI
Government
Federal agencies, ministries, and public institutions deploying AI in citizen-facing services. I developed the Trust by Design framework through research across 14+ AI implementation teams in the German public sector.
Healthcare
Hospitals, insurers, and health technology companies using AI for diagnosis, triage, or patient decisions — among the highest-risk applications under the Act.
Finance
Banks, insurers, and financial institutions using AI for credit, fraud, or eligibility decisions. Automated decisions affecting individuals carry full high-risk obligations.
The Research Behind the Framework
Built inside government AI. Not consulted from the outside.
The Trust by Design framework was developed through independent research across 14+ AI implementation teams in the German public sector — the same environment where the EU AI Act's requirements will be most intensely felt.
That research was validated through advisory sessions with process management and CX leaders in regulated government institutions, and presented at the Global Digital Transformation & CX Summit. The MA thesis underpinning the framework was presented to a large internal audience at Bundesdruckerei, one of Germany's most security-critical public sector organisations.
The framework addresses the specific patterns of failure I observed in real deployments: AI systems that were technically sound but humanly unusable, inaccessible to the citizens they were built to serve, and structurally unable to meet human oversight requirements — even when teams believed they had addressed them.
14+
AI implementation teams researched
MA
Design for Responsible AI, ELISAVA
EU AI Act
specialist since 2022
Get Started
Discovery calls are 30 minutes. No obligation.
Tell me about your AI system, your deployment timeline, and where compliance feels most uncertain. I'll give you an honest assessment of where the human layer gaps are and what it would take to close them.