Meta and the EU code of practice — what the dispute among large providers tells companies in Central Europe

What the dispute was about

The code of practice is a voluntary instrument that helps providers of general-purpose AI models demonstrate compliance with the AI Act. A provider that has not signed it must still meet the obligations of the regulation, but demonstrates this in another way. Refusing to sign therefore does not exempt anyone from the law; it changes the way the provider deals with the regulator and the degree of predictability for its customers.

What the following months showed

In 2026 the EU amended the AI Act by Regulation (EU) 2026/1744, known as the AI Omnibus. It moved the obligations for high-risk systems to 2 December 2027 and 2 August 2028 and softened the wording of Art. 4 on AI literacy. The transparency obligations of Art. 50 have applied since 2 August 2026; the transition period until 2 December 2026 concerns only machine-readable marking of content by providers of tools placed on the market before 2 August 2026. In our assessment, the EU was responding to arguments about costs and legal uncertainty, but it did not retreat from the core of the regulation.

In the first version of this text we argued that the AI Act might push sensitive innovation out of the EU and reduce European companies to implementing other people’s technology. That risk is still disputed among economists and cannot be settled today. More useful for the board is a practical question: how should the company operate when its providers approach European rules in different ways?

What this means for companies in Central Europe

• The question to the provider is part of the purchase. When choosing a tool, it is worth asking how the model provider demonstrates compliance with the AI Act and what it makes available to customers.
• The process description must not be tied to one model. If a provider changes its terms in Europe, the company should be able to replace the model without losing the work put into describing its processes.
• Practical sector guidance is still scarce. Construction and industrial companies ask whether their uses fall into the high-risk category and how to document decisions supported by algorithms. These questions call for assessment of specific cases, not general statements.

The last point is why it is worth taking part in shaping good industry practice instead of waiting for ready-made answers. This is the role the Foundation plays in semantic transformation: organising a company’s names, processes and data so that people, language models and board reports understand them the same way.

What has changed since first publication

• We corrected the description of the event: the original version conflated the refusal to sign the code of practice for general-purpose AI models (July 2025) with a different, earlier document on disinformation.
• We removed figures on the EU’s share of global AI investment taken from a press release without a methodology.
• We added the effects of the AI Omnibus (Regulation (EU) 2026/1744) and the applicable timetable.
• We replaced the thesis of “Europe drifting towards the periphery” with questions the board can settle today.