Remember when I called the EU AI Act a self-inflicted wound that would turn Europe into a digital colony? Well, the Commission just proved me right by proposing to delay high-risk AI obligations by a year and water down several key requirements. They’re calling it the “Digital Omnibus on AI regulation proposal.” I call it what it is: panic.
Too Little, Too Late
Let’s be clear about what’s happening here. The European Commission isn’t softening regulations because they’ve suddenly discovered the virtues of innovation. They’re doing it because the entire framework is collapsing under its own bureaucratic weight.
The proposed changes read like a confession of failure:
High-risk AI obligations delayed by 12-16 months: Annex III systems get until December 2027, Annex I systems until August 2028. Why? Because after all their grandiose planning, they still haven’t figured out the technical standards companies need to comply with. They literally regulated an industry before knowing how to regulate it.
AI literacy requirements gutted: What was once a binding obligation on providers and deployers is now just “encouragement” from the Commission and Member States. Translation: unenforceable wishful thinking.
Registration requirements relaxed: Companies no longer need to publicly justify why their systems aren’t high-risk under Article 6(3). Less transparency, more wiggle room.
Penalty exemptions expanded: The Commission is extending reduced penalty ceilings, previously reserved for small and medium enterprises, to small mid-cap companies. More businesses get lighter consequences, which certainly sends a message about enforcement priorities.
The Acceleration Clause: Having Your Cake and Regulating It Too
Here’s the truly European touch: the deadlines aren’t actually fixed. Instead, high-risk obligations kick in once the Commission confirms that ‘harmonised standards, common specifications or Commission guidelines’ are available: six months later for some systems, twelve for others. The problem? These standards were supposed to be ready by 2025. Standardization bodies now say 2026. Who’s to say they won’t slip again?
So companies face a moving target tied not to calendar dates, but to the Commission’s assessment of when bureaucratic infrastructure is finally ready. They’re admitting the original timeline was based on standards that don’t exist yet, while making companies wait for regulators to finish their homework before knowing when compliance actually begins. This is regulatory theater at its finest.
Meanwhile, in the Real World...
While Europe debates whether to delay its self-destruction by 12 or 16 months, here’s what’s actually happening:
DeepSeekjust dropped models that rival GPT-4 class performance at a fraction of the cost.
Metais preparing Llama 4, which Europeans might not even have full access to.
OpenAIandAnthropicare racing toward AGI while Europe argues about documentation requirements.
China, the UAE, and Singaporeare building massive AI infrastructure without asking permission from bureaucrats
The delay doesn’t change the fundamental equation. Europe still has:
No significant compute infrastructure.
No competitive semiconductor capabilities.
Energy prices 2.5x higher than the US.
A regulatory mindset that treats innovation as a threat rather than an opportunity.
The Brain Drain Accelerates
You know what these “relaxed” regulations tell AI entrepreneurs and researchers? Europe might strangle innovation slightly slower than originally planned. That’s not exactly a compelling pitch.
The talent exodus I predicted is already happening. Top researchers are heading to labs in the US, China, and the Gulf states. Startups are incorporated in Delaware and Texas, not Dublin. The best and brightest see these regulatory adjustments for what they are: deck chairs on the Titanic.
A Step in the Right Direction?
Is this progress? Sure, in the same way that running toward a cliff slightly slower is progress. The fundamental problem remains: Europe is trying to regulate its way to AI leadership while everyone else is building their way there.
These changes reveal the EU’s central delusion: that you can have comprehensive, precautionary AI regulation AND competitive AI development. You can’t. Every month spent debating compliance frameworks is a month where American and Chinese labs pull further ahead. Every euro spent on regulatory overhead is a euro not spent on GPUs.
The Commission’s proposal is an admission that its grand regulatory edifice was built on sand. But instead of tearing it down and starting over with an innovation-first approach, they’re just pushing the collapse date back by a year.
The Clock Is Still Ticking
Here’s the brutal truth: these delays change nothing fundamental. Europe is still bringing a regulatory framework to an infrastructure fight. While Brussels debates timelines, Silicon Valley and Shenzhen are shipping products. While the EU softens literacy requirements, the rest of the world is achieving AI literacy by actually building and deploying AI systems.
The tragedy isn’t that Europe chose wrong. It’s that even when faced with obvious failure, they can only conceive of marginal adjustments rather than fundamental change. The AI revolution won’t wait for European bureaucrats to finish their paperwork.
Wake me when Europe announces they’re scrapping the AI Act entirely and launching a Manhattan Project for AGI. Until then, this is just rearranging deck chairs while the band plays on.
The race continues. Europe just announced it’s switching from cement shoes to slightly lighter concrete ones.
VAE VICTIS.