The UplinkSubscribe
The EU AI Act: A Compliance Challenge with Dubious Benefits

The EU AI Act: A Compliance Challenge with Dubious Benefits

The EU AI Act will impose heavy compliance costs on companies while delivering little in the way of clear benefits. From Politechly.

User avatar placeholder

January 31, 2025

EU AI Act compliance will impose heavy costs on companies while delivering little in the way of clear benefits. I made that case on Politechly in January 2025, days before the law’s first provisions took effect on February 2. Those provisions banned the AI practices the law treats as posing an unacceptable risk and required companies that build or use AI to take measures to ensure, to their best extent, “a sufficient level of AI literacy” among their staff.

The law runs about 90,000 words, and even the Future of Life Institute’s high-level summary tops 2,000. A small company can’t interpret all of that on its own, so it pays law firms and consultancies to do the reading. Those firms are publishing guidance that concedes they aren’t sure what terms like “emotion recognition”, or the literacy standard itself, mean either.

The Act also reaches well past AI developers to cover any company using AI in the European Union. An American firm running a customer chatbot or an internal HR tool has to work out which of the law’s duties apply to that tool, including prohibitions the law itself never clearly defined. A wrong guess about a banned practice can cost a large company up to 7 percent of global revenue.

The AI Act follows the script Europe wrote with GDPR, its privacy law, which built a compliance industry without preventing the major privacy scandals it was sold to stop. The largest companies can best afford to comply, and they’re already shaping the industry’s standards, which forces startups that can’t carry the costs to slow down or exit the European market.

In July 2026 the EU published the AI Omnibus, which softened the literacy duty to “support the development of AI literacy.” It also pushed back the rules for systems the law classes as high-risk, to December 2027 for some and August 2028 for others.

Brussels should take the approach I’ve urged on California and in AI Regulation Needs a Light Hand: drop the sweeping mandates and work with industry on targeted rules aimed at demonstrated harms, with guidance a company can follow without retaining a law firm.

The full piece, with its sources and charts, is on Politechly: The EU AI Act: A Compliance Challenge with Dubious Benefits.

This article was produced in part with AI, see my disclosure here.

Image placeholder

Gregory S. McNeal

Founder and Editor-in-Chief of The Uplink.