The “AI Act” will allow AI Developers in the EU to Regulate Themselves without Consequences

The European Union is poised to pass the “AI Act”, a broad legislative attempt to regulate AI in the name of preventing its potential harm. Recent negotiations over the details represent a victory for developers of the most powerful AI models.

The legislation has faced fierce opposition from France, Germany, and Italy over concerns about how “Foundation” models would be regulated. (Foundation models are those which are trained on large amounts of data and are capable of performing a diverse array of tasks, such as creating various types of media, engaging in human-like conversations, and generating computer code.) In the original proposal, the AI Act would impose tighter regulation on these Foundation models than on less-capable AI, even if the Foundation models were to be used for a particular, narrowly defined purpose. This fact raised concerns that the regulation would place Europe’s big AI players at a competitive disadvantage relative to their American counterparts. France was particularly concerned, believing that Mistral could match or exceed the capabilities of ChatGPT in the near future if not hampered by legislative restrictions.

Reuters is reporting that France, Germany, and Italy have gotten their way: Foundation models will be regulated, but will not be subject to stricter regulation than other models.

So, what’s in store for creators of Foundation AI models under the new agreement? Self-regulation without consequences.

Under the new proposal, creators of AI models (including Foundation models) will regulate themselves. Its not yet clear what this regulation will entail beyond a requirement that AI developers will have to create and disseminate “cards”. Among other things, these “cards” will explain the capabilities and limits of the model, provide information about its intended uses, and report the results of studies conducted on its level of bias. At least for now, there will be no sanctions for those who do not comply with the mandatory self-regulation, though the door has been left open for a future sanctioning body to be created in future.