EU AI Act: Myth and Reality

MashMachine
Podcast
Law & Legislation
#DigitaleWissensbissenS01E12

The European AI Regulation (EU AI Act) has been law since August 1, 2024. And as with any major regulation, two camps have suddenly emerged: Some see it as the end of all innovation, while others see it as finally bringing order to the unchecked growth of AI. In this episode of “Digital Bites of Knowledge”, we break down the topic—in a practical way, with a focus on companies that don’t build AI, but rather use it.

It quickly becomes clear: The law is not some abstract “Brussels thing” for OpenAI & Co. It also affects ordinary companies—at the latest when AI is running productively and influencing real decisions.

Key Quotes

  • This is applicable law. The only thing that’s phased in is the implementation.”
  • “Many people think the AI Act only applies to large AI developers – that is completely wrong.”
  • “The AI Act distinguishes between providers and deployers – and the deployer has its own obligations.”
  • “The only decisive question is: Does the system evaluate or select natural persons?”
  • “Transparency sounds trivial—but in practice, I regularly see corporate chatbots where this isn’t communicated.”
  • AI literacy isn’t just a nice recommendation, but a standalone obligation.”
  • “Just like with the GDPR: You can wait it out—but later it will be more expensive and hectic.”

Summary

The EU AI Act is not just a set of rules for model providers, but also affects companies that use AI in their processes—especially where systems evaluate, rank, or prepare decisions involving people (e.g., in certain HR functions or when granting access to specific services). What matters less is the buzzword “AI” and more the specific use case: Does the system merely provide support (e.g., a chatbot), or does it play a role in decision-making?


The episode also outlines a clear course of action for companies: Don’t panic—but don’t wait until regulatory frameworks are fully in place either. Those who get started now will gain a head start through inventory, classification, vendor due diligence, documented oversight, and training. The AI Act doesn’t make AI impossible—but “just deploying it quickly” is becoming an increasing risk.

MashMachine
MashMachine
AI servant
Artificial intelligence that multiplexes your efforts.

More blog posts

260 Chicken McNuggets: Where AI Pilot Projects Hit the Wall

Why do 95% of AI pilots deliver nothing measurable? Because the demo shows the system's best day — while production gets judged on its average. Episode 2 dissects the most expensive stretch of any AI project: the one between the applause for the demo and the Monday morning it actually has to run. With real cases — Klarna, McDonald's and its 260 Chicken McNuggets, the sobering IBM and MIT numbers — and the one question that separates the salespeople from the builders.

The smarter the AI, the subtler the hallucinations

Why does AI invent sources, quotes, even entire court rulings? Because it doesn't lie — it guesses, and we trained it to. The false sentence sounds exactly like the true one, and the better the models get, the harder it is to catch. Episode 1 uses real cases — Deloitte, Air Canada, a lawyer fined $10,000 — to show why hallucinations aren't a bug but a feature. And the one shift that turns AI from a liability into a tool you can build a process on.

Artificial intelligence - technology or strategy?

AI is everywhere—but it’s rarely effective in day-to-day business operations. This episode highlights three common pitfalls: a wait-and-see approach (which leads to shadow AI), a “co-pilot for everyone” strategy with no real impact, and a Center of Excellence disconnected from real-world problems. Instead of treating AI as an ideology, the key is to strike a balance: digitize processes, identify bottlenecks, and use AI to significantly reduce the workload on top performers and scale operations.