The AI Act’s strong enforcement powers are set to kick in on 2 August
In just a few days, the EU will be able to start using the key tools in its AI rulebook. The question is whether or not the bloc will dare to do so.
Two years ago, the EU signed off on the AI Act, a set of comprehensive rules that put obligations on AI models and systems according to perceived risk. Outside the law’s few banned uses of AI, the most stringent requirements apply to the largest general-purpose models trained by giants like OpenAI and Anthropic.
When the law was being drafted, so-called “systemic” risks – such as AIs enabling large-scale cyberattacks – seemed far-off. But the recent advent of cyber models like Anthropic’s Mythos has made such threats much more tangible.
At the same time, the question of who gets access to frontier AI models has become a matter of politics.
