Brussels Softens the AI Act. The IMF Says the Risks Are Growing.

EU lawmakers ease AI regulation as the IMF warns of mounting AI-driven financial risks.

EU lawmakers agreed today to weaken the world's first comprehensive AI regulation, just as the IMF warned that artificial intelligence is amplifying cyberattacks and threatening financial stability.

The provisional deal, struck after months of intense lobbying from US tech giants and European startups, eases two of the AI Act's most contentious provisions. Open-source AI models, including Meta's Llama and Mistral's family of systems, will face significantly lighter compliance obligations. The threshold for classifying AI systems as "high-risk," which triggers the law's most burdensome requirements, has been raised. Fewer systems will now qualify. Implementation timelines are expected to be extended.

The changes amount to a meaningful retreat from the original law's ambitions. Companies that invested heavily in compliance infrastructure for the stricter version may find they over-built. Enterprise AI teams across Europe now face a different calculus: compliance roadmaps drafted over the past 18 months need revisiting, and competitive dynamics with US and Chinese rivals have shifted.

The lobbying argument was straightforward. Europe's original rules would have handicapped its own AI sector while Washington and Beijing moved faster with fewer constraints. Brussels, it appears, agreed.

But the timing is sharp. On the same day, the IMF published a warning that AI-fuelled cyberattacks are creating mounting risks to global financial stability. The fund's concern is not hypothetical: AI is already being used to scale phishing operations, manipulate markets, and exploit vulnerabilities in financial infrastructure faster than institutions can respond.

The two stories sit in direct tension. One says the rules were too heavy. The other says the risks are accelerating. For decision-makers in enterprise tech and financial services, both are true simultaneously, and that is the operating reality.

No Polymarket prediction market currently tracks EU AI regulation, so this is a real-world news brief without a market signal.

The next watchpoint is the European Parliament's formal ratification vote, expected before summer recess. The softened text still needs to survive that process. If it does, Europe's AI regulatory framework will look substantially different from what was signed into law.