The EU has spent four years writing rules for artificial intelligence. Now it’s building the machinery to actually enforce them.

Last week, Brussels appointed a 60-member Scientific Panel and a 174-member Advisory Forum to support the AI Office and national regulators as the Act moves into its main enforcement phase on August 2, 2026. The panels include experts in frontier AI, engineering, technical auditing, and societal impact. Permanent participants include the EU Agency for Fundamental Rights and ENISA, the EU’s cybersecurity agency.

On paper, this looks like robust governance. In practice, it’s the beginning of the hard part.

Why the Expert Layer Matters

The AI Act is built on risk classification: high-risk systems get strict requirements, lower-risk ones get lighter touch. But risk classification is not a simple technical exercise. A model that looks innocuous in one context — say, an AI that ranks job applicants — becomes high-risk when it’s making consequential decisions about people’s livelihoods.

The Scientific Panel’s mandate is to help regulators make those judgement calls. They’ll advise on model classification, evaluation methods, and whether systems that evolve after deployment need closer scrutiny. That’s genuinely difficult work. It requires people who understand both the technology and its real-world effects.

The Advisory Forum brings in civil society, which is the right instinct. AI governance can’t be reduced to code audits. Questions about discrimination, privacy, worker protection, and democratic accountability often emerge at the point where technical systems meet actual people. Having those voices in the room — not just industry — matters.

The Political Test Nobody’s Talking About

Here’s what will actually stress-test this system: general-purpose AI models are mostly developed outside Europe, by companies with significant political resources. When the Scientific Panel produces findings that are inconvenient for a major AI lab — classifying a powerful model as higher-risk than the company would like, or concluding that evaluation methods are insufficient — there will be pressure. Political pressure. Legal challenges. PR campaigns.

The question is whether the expert architecture can withstand that. Advisory bodies only work if their advice is transparent enough to build public trust, and if the institutions backing them have the spine to act on unwelcome findings.

The Commission has given these bodies two-year terms. That’s sensible for continuity, but it also means they’re not a permanent fixture — another variable in a system that’s already complex.

What This Actually Means for Businesses

If you’re deploying AI in Europe, the expert panels are indirectly your problem. Their technical assessments will shape how regulators interpret compliance obligations. A finding that current evaluation methods are inadequate for certain model types could mean additional testing requirements. Recommendations on systemic risk could trigger closer scrutiny for models used across multiple sectors.

The August 2026 deadline is real. Most of the AI Act’s requirements apply from that date. Companies that haven’t been paying attention to the enforcement architecture are about to find out that “we didn’t know” is not a compliance defence.

The Bottom Line

Europe has built something genuinely novel here: a multi-disciplinary expert layer between AI companies and political decision-makers. Whether it makes AI safer in practice depends entirely on whether it can operate independently under pressure.

The legislation was the easy part. Now the real test begins.

Source: European Times