The EU AI Act: From Ethical Principles to Enforceable Corporate Responsibility

On 1 August the EU Artificial Intelligence Act entered into force. Three years after the Commission’s proposal, which I wrote about in 2021, Europe has the world’s first comprehensive horizontal AI law.
The final text is broader than the original proposal. The arrival of ChatGPT during the negotiations forced legislators to add a separate regime for general purpose AI models, with additional obligations for the most capable ones. The core logic, however, remains the risk based approach.
The obligations arrive in stages. Prohibited practices apply from February 2025. Rules for general purpose AI models apply from August 2025. Most requirements for high risk systems apply from August 2026. The most serious violations can lead to fines of up to 35 million euros or seven percent of global annual turnover.
Not only for technology companies
The most common misunderstanding I hear is that the AI Act is a problem for AI developers. It is not only that. The regulation distinguishes between providers, who develop or place AI systems on the market, and deployers, who use them in their operations. Deployers have obligations of their own, particularly for high risk systems: using them as instructed, ensuring human oversight by competent people, monitoring their operation and, in some cases, assessing the impact on fundamental rights.
A bank using a purchased credit scoring model, a manufacturer using an AI tool to screen job applicants, an insurer using an external fraud detection system: all of them are inside the scope.
From statement to operating model
Until now, responsible AI in most organizations has been a set of principles and, in better cases, a review committee. The AI Act turns it into something that needs an operating model.
Inventory. A complete register of AI systems in use, including those embedded in software bought from vendors.
Classification. A consistent method for determining which systems fall into which risk category.
Ownership. Named business owners for each material system, with clear accountability.
Controls. Documentation, data governance, human oversight and monitoring built into the lifecycle.
Supplier management. Contracts and due diligence that ensure vendors provide what deployers need to meet their own obligations.
Literacy. The Act explicitly requires organizations to ensure that staff dealing with AI have a sufficient level of AI literacy.
Compliance or capability
There are two ways to respond. One is a legal exercise: interpret the minimum requirements, produce the documentation, pass the audit. The other is to use the regulation as a forcing mechanism to build what most organizations needed anyway: visibility over their AI systems, clear ownership and a way to know when something is going wrong.
The second approach costs more initially. It also produces something valuable. Organizations that know which AI they use, where and with what results will scale AI faster than those still discovering their systems during an audit.
For private equity, the AI Act is now a due diligence topic. A target company that relies on AI in high risk areas without governance is carrying a regulatory liability that should affect valuation and the value creation plan.
Leadership question: Is AI compliance becoming a legal exercise, or should it be an opportunity to build a more trustworthy and competitive organization?


