Skip to content
Articles
ArticlesAI translation· 9 July 2025

Episode 10.8 - No risk, no fun with AI? What does the law have to say?

by 5 min read

Quite a lot of "AI Risks & Failures" stories have been landing on my news dashboard lately, so I've moved this topic up the queue. Mainly because the legislator has already set 2 deadlines for it: 2 February 2025 (already passed) and 2 August 2025 (so, in 3 weeks).

The EU AI Regulation (Artificial Intelligence Act)* entered into force on 1 August 2024, and it hits companies using artificial intelligence with a broad set of new compliance requirements. The law takes a risk-based approach and distinguishes between the various players along the AI value chain. For now, I'll only cover the points that affect professional AI users.

The basic obligations for every company using AI

1. Build AI "User Awareness / AI Literacy"

The first obligations of the AI Act have applied since 2 February 2025. All providers and deployers of AI systems must ensure that their staff, and anyone else involved in operating and using AI systems, have a sufficient level of AI competence (AI Literacy). This obligation applies regardless of the AI system's risk class and therefore affects almost every company that uses AI.

AI competence covers the skills, knowledge and understanding needed to use AI systems in an informed way and to be aware of their opportunities and risks. Put simply: employees should use AI systems competently, but they absolutely must recognise both the potential and the risks. That means not just the technical knowledge itself, but also the ethical, social and legal consequences. Companies have to put appropriate training measures in place, taking into account technical knowledge, experience and the specific context of use. Important: this obligation applies to every business, regardless of size!

You'll also need training concepts tailored to each target group:

  • Developers: detailed technical and ethical requirements. If they work on their own models or on integrating them, expect additional obligations here too, since this can turn you into a "downstream provider" (see below).
  • Managers: compliance, accountability and risk management
  • Users: basic training on safe AI use
  • Data protection / compliance officers: specialised legal training

2. Build AI governance, compliance and risk management

Companies using general-purpose AI models** (General Purpose AI, GPAI, i.e. ChatGPT, Copilot, Midjourney and similar tools) are subject to special provisions from 2 August 2025. While the main responsibility lies with the AI providers, companies using these tools must check whether the way they integrate them turns them into downstream providers (which brings additional obligations that I won't list here for now).

Among other things, the regulation specifically requires companies to:

  • disclose when such systems are in use,
  • take technical and organisational measures to prevent misuse,
  • keep documentation of the contexts in which AI is used,
  • and, where necessary, ensure human oversight.

A critical step here is a complete inventory of all AI systems used in the company. That covers both in-house and externally sourced AI applications, including standard software with AI features (e.g. Microsoft Copilot, Google Gemini etc.) that arrived via updates. Companies should also think about "shadow AI": AI tools employees use on their own without the responsible department knowing about it!

In many cases, using AI in the workplace touches on the co-determination rights of the works council. Because as soon as AI systems are used that affect employees' behaviour, performance or decisions about them, the works council should be involved.

That's quite a bit of effort, and in my view it's entirely justified, because AI tools are pretty disruptive. Alongside the benefits, they bring a number of new kinds of risk that classic compliance and risk management frameworks simply don't cover yet.

Deploying AI is a management job!

Right now I'm once again seeing a familiar tendency among some decision-makers: "they're IT tools, so let IT deal with it." No. In this case it's definitely not a technical task. Top management has to engage with this directly, and organisational structures need to be built to steer AI governance. For example, the following positions and roles could be defined:

  • Chief AI Officer (CAIO) for strategic leadership of AI initiatives
  • AI compliance officers as a central coordination point
  • Interdisciplinary AI board with representatives from legal, the works council, IT, data protection, business units etc ...
  • AI Ethics Committee for ethical assessments

I hate repeating myself, but you mustn't fall into the "Simplicity Trap", and from 2 August that comes with legal consequences too.

It would also be wrong to play it safe because of the risks and put off evaluating and introducing AI tools. On the other side of the coin, this technological innovation is very powerful and, when used properly, it delivers tangible and fast competitive advantages.

Well, as the saying goes: “With great [AI] power comes great responsibility”*** 😉


*More on this topic can be found at the AI Service Desk of the Bundesnetzagentur


** There are further risk classes for AI systems that are subject to considerably stricter requirements.


Live long and prosper 😉🖖


*** rusic meference ;)


Postscript: I'm not actually a big fan of regulation, especially when it's badly done. But I have to say the EU is doing a good job when it comes to digital law and the digital society. I already mentioned the "Digital Competence Framework for Citizens" in 10.1. With data protection and the "Digital Market Act", the EU is trying to curb the market dominance and shady practices of "Big Tech USA", and on AI it has chosen a proactive approach that prevents without blocking.

Read on
Topic
AI Governance
▶Listen