How a Panel of Experts Shaped the EU's Blueprint for Trustworthy AI

When people worry about artificial intelligence making decisions about their lives, whether it's a hiring algorithm, a loan application, or a medical diagnostic tool, they're really asking one question: can I trust this system to treat me fairly? That question sat at the heart of a quiet but consequential chapter in European policymaking, one that began years before the EU's AI Act became law.

The European Commission didn't start by writing rules. It started by asking a group of experts what trustworthy AI should even mean. In 2018, the Commission appointed the High-Level Expert Group on Artificial Intelligence, known as the AI HLEG, to advise on the bloc's broader AI strategy. Think of it like consulting a panel of engineers and ethicists before building a bridge, you want to know the load-bearing principles before anyone pours concrete.

During its first year, the group produced its foundational document: a human-centric approach to AI built around seven key requirements that systems must meet to be considered trustworthy. These weren't abstract ideals. They became the scaffolding for everything that followed, including the eventual legal framework now shaping how companies deploy AI across the continent.

The group didn't stop there. Building on that first deliverable, it issued 33 recommendations aimed at steering trustworthy AI toward sustainability, growth, competitiveness, and inclusion. The goal was twofold: empower European citizens while also protecting them. That balance, between innovation and safeguard, has remained the defining tension in EU tech policy ever since.

From Guidelines to a Working Tool

The group presented its findings at the first European AI Alliance Assembly in June 2019. That event mattered enough that the Commission extended the group's mandate by another year, giving it room to test its ideas in the real world rather than leave them as theory.

That extended mandate led to a pilot of the Ethics Guidelines for Trustworthy AI, essentially a trial run to see if the principles held up outside a conference room. When the group's mandate ended in July 2020, it left behind two more practical deliverables.

The first was a self-assessment checklist, a tool that translates abstract ethical guidelines into something a developer can actually use day to day. It exists as both a web-based prototype and a downloadable PDF. Picture it as a pre-flight checklist for engineers, a way to verify that fairness, transparency, and safety considerations have been addressed before a system goes live.

The second explored how those 33 recommendations might work in three real-world sectors: the public sector, healthcare, and manufacturing paired with the Internet of Things. These are precisely the areas where AI's promise and its risks collide most directly. A hospital deploying a diagnostic algorithm faces very different stakes than a factory floor optimizing supply chains, and the group's work tried to account for that.

None of this happened in isolation. The AI HLEG worked closely with the European AI Alliance, an online forum with more than 4,000 members spanning academia, business, civil society, and policymakers. That's a sizable crowd for a policy consultation, and it mattered. Alliance members gave detailed feedback on the Ethics Guidelines and contributed policy papers and academic research that shaped the group's other outputs.

The collaboration culminated in two assemblies. The first, in June 2019, drew 500 forum members together in person to engage directly with the Commission's policymaking process. The second, held online in October 2020 amid the pandemic, brought in more than 1,900 participants to discuss the results of the public consultation on the Commission's White Paper on Artificial Intelligence. Even after the expert group's formal mandate ended, the Alliance kept meeting and debating.

The influence of this early work is hard to overstate. The concept of trustworthiness and the seven key requirements from the Ethics Guidelines now guide the EU's legislative approach to AI. The group's recommendations fed directly into the Communication on Building Trust in Human Centric Artificial Intelligence, the White Paper on Artificial Intelligence: a European Approach to Excellence and Trust, and the updated Coordinated Plan on AI. Those documents, in turn, laid groundwork for the binding rules that followed.

More recently, the Commission has kept this advisory infrastructure active. In October 2026, it convened a special meeting of the Scientific Panel on frontier AI safety and risks, a sign that expert consultation on AI's sharpest edges hasn't gone away just because the original group's mandate expired. The practice of bringing in outside expertise before legislating, rather than after, has become something of a template.

Why It Matters

Policy built without expert input tends to either overreach or miss the mark entirely. Regulators who don't understand how a technology actually works risk writing rules that sound tough but fail in practice, or rules so vague that bad actors slip through anyway. The AI HLEG's work offers a counterexample: a sustained, multi-year effort to ground policy in both ethical reasoning and technical reality before a single enforceable law was drafted.

For ordinary citizens, the payoff isn't abstract. The seven trustworthiness requirements that emerged from this process now underpin protections written into EU law, covering things like transparency about when you're interacting with an AI system and accountability when something goes wrong. The self-assessment checklist gives smaller companies, not just tech giants, a usable starting point for compliance. And the continued activity of the AI Alliance, along with newer bodies like the Scientific Panel on frontier AI safety, suggests Brussels still sees value in keeping outside experts at the table long after the original mandate expired. That's a model worth watching as other governments grapple with the same questions.