Article icon
Article

GDPR Déjà Vu: Don’t Repeat the Same Mistakes with the AI Act

Europe is once again leading the way on a first-of-its-kind tech regulation. And, in a replay of years past, teams are sleepwalking into a compliance nightmare.

Roughly a decade after the General Data Protection Regulation (GDPR), the AI Act promises many of the same headaches. That’s because meeting the rules isn’t just about rewriting data processes and systems but also troubleshooting, training, and tracking.

Even though complying with the world’s first automation regulation recently got pushed back to December 2027, teams shouldn’t make the same mistake of thinking there’s enough time (until there isn’t).

We’ve Seen This Film Before

I’ve previously written that American enterprises should learn from the mistakes made by European companies under the GDPR. In that regulation, teams too often underestimated what was required to bring the data protection rules from paper to practice, particularly in staff training, data mapping, and recordkeeping.

For example, in the months leading up to GDPR compliance in 2018, one-third of EU companies were concerned their stack couldn’t effectively manage data in line with the new code. Those fears were well-founded with the average firm spending €1.3 million just to prepare for GDPR. Worse, compliance gaps arose regardless (with knock-on reputational effects). I’m worried the same thing is happening again, but this time for automation.

At its core, the AI Act is about ensuring human oversight remains front and center in the intelligence revolution. The bloc created a risk-based approach to classify what is and isn’t acceptable, with the highest tier already prohibiting practices such as social scoring and emotion recognition. The next tier, high-risk systems, applies to AI that “can pose serious risks to health, safety, or fundamental rights” across biometrics, critical infrastructure, employment decisions, essential services, and more.

This is a broad category in what it mandates and who it impacts. Providers – those who build AI systems – must document compliance across training data, technical records, and risk management. Deployers, the companies using AI day to day, also must assign oversight personnel, monitor performance, and report serious incidents to providers and regulators. The headline is that you don’t have to have written a single line of code to be on the hook for a potential multimillion-euro fine.

Don’t Fall for the Same Traps

Like GDPR before it, this regulation covers much more than companies likely realize. Even the bloc appears to recognize this and essentially admits it needs more time to actually enforce these wide-ranging rules. Over the summer, with guidance still being finalized and many member states yet to appoint national regulators, the bloc pushed high-risk AI compliance back from August 2026 to December 2027. 

The runway is now longer, but companies can’t afford to be complacent. Instead, AI providers and deployers alike need to use this time to their advantage.

First, creating the newly mandated “human in the middle” will take money and time. It’s not enough to assign someone – organizations must document that humans are genuinely reviewing automated decisions at each stage and have auditable records to prove it. This is an operational project that most teams haven’t started.

Second, most don’t have a clear inventory of AI. It’s common sense, but you can’t govern what you can’t see. Here, teams need to map out their ecosystems and truly understand what automation touches. Then work backward to establish compliance documentation and human oversight before regulators come knocking. In most organizations, data teams are best placed to lead this. Article 10 of the AI Act requires high-risk systems to be trained on documented, high-quality data, which is work they already do, day in and day out.

Third, international companies need to consider their liability. Enterprises that offer AI systems in the EU, or whose AI outputs are used there, are in scope. For most significant American enterprises, that’s not a hypothetical. Again, this is something we saw with GDPR. Meta learned this the hard way after transferring European user data to the United States, resulting in a €1.2 billion fine in 2023.

Be aware that the AI Act also asks where a system’s output ends up. A model hosted in Virginia that screens candidates for a Dublin office is in scope, even if this kind of system isn’t immediately apparent as “European.” The data inventories companies built for GDPR are a useful starting point, but they won’t show which models produce decisions that cross borders, or who acts on them.

The Time to Begin Compliance Was Yesterday

These are technical and cultural conversations that teams need to have from the top down. 

Every level of the company will need to be more careful about unauthorized automation (shadow AI) and this includes the C-suite. More than two-thirds of corporate executives admit to violating their own AI usage policies – the same leaders who rank security and compliance as their greatest implementation challenge. This world-first regulation only highlights the gaps and the work ahead to fill them. 

Remember that the bloc is a proven bellwether, and more than 100 countries followed its lead on data privacy. This is also true of states like California and Colorado, even if federal data privacy is still on the horizon. Right now, the same regulator is rewriting the rules on AI. American companies should be careful not to dismiss this as a European problem. Instead, AI providers and deployers around the globe need to think objectively about where regulation is headed and get ahead of it.  

For the moment, however, an extra year for high-risk systems to get up to code is a lifeline. Use this time to take stock of AI, bake in the human final say, establish an inventory, and document oversight chains.

After all, beyond merely following the rules, these are the kind of baselines that safe and proper automation deserves. As history shows, regulatory compliance is never one-and-done, and complacency is only ever a liability.

Build your AI governance skills in 2026.

Explore training programs on AI governance, data governance, and compliance for data practitioners.