ChatGPT broke the EU's plan to regulate AI

ChatGPT didn’t just captivate millions of internet users over the past few months—it also completely broke the European Union’s playbook for artificial intelligence.

Two German lawmakers recently delivered speeches on the European Parliament floor that were drafted entirely by ChatGPT. The stunt was meant to send a clear message: lawmakers need to get a grip on AI. The awkward truth, however, is that Brussels is currently scrambling to figure out how to do that without wrecking its own signature rulebook.

An old blueprint meets a general-purpose engine

Back in 2021, the European Commission proposed the landmark Artificial Intelligence Act. The original plan was simple: regulate AI based on its specific use case. Unacceptable applications like social scoring or certain types of facial recognition would be banned outright, while sensitive uses in education, hiring, and law enforcement would be labeled "high-risk," triggering strict demands for human oversight and transparency.

Then ChatGPT showed up and shattered that entire taxonomy.

Large language models are general-purpose engines. They can write poetry and computer code, but they can also churn out mass disinformation, write court rulings as a Colombian judge recently admitted, or threaten researchers as Bing’s AI chatbot did during testing. As Romanian lawmaker Dragoș Tudorache pointed out, these systems are powerful engines that arrive without a predetermined purpose.

Scrambling for an answer, Tudorache and co-rapporteur Brando Benifei proposed in February that AI systems generating complex text without human supervision should be added to the "high-risk" list. That immediately sparked pushback. Prominent center-right lawmaker Axel Voss argued the move would inadvertently brand harmless daily software as dangerous. Meanwhile, policy advocates like Mark Brakel from the Future of Life Institute warned that targeting only text-based tools leaves other risky generative models off the hook.

The lobbying blitz behind closed doors

With three-way negotiations between the Commission, Council, and Parliament expected to kick off as early as April, general-purpose AI is rapidly turning into a major deadlock.

Big Tech is watching every move. According to an investigation by Corporate Europe Observatory, giants like Microsoft and Google have been aggressively lobbying EU officials to keep general-purpose foundation models out of the high-risk bracket. Microsoft’s Chief Responsible AI Officer Natasha Crampton publicly urged the EU to stick to narrow, high-risk use cases, arguing that drafting emails and writing code are low-risk activities that Europeans shouldn’t lose access to.

In a twist of irony, when Politico asked ChatGPT whether it should be regulated, the chatbot itself replied that the EU should consider classifying generative AI and LLMs as "high risk" due to their potential to produce harmful and misleading content.

Why it matters

Brussels has run straight into the core paradox of frontier AI: traditional governance models rely on knowing what a product is built for, but generative models can do almost anything.

If Parliament succeeds in slapping a broad "high-risk" label on complex text generators, it could significantly raise compliance barriers across the continent. While tech giants like Microsoft and Google have the legal muscle to adapt, smaller European startups and open-source builders might find themselves crushed under administrative burdens, potentially creating an unintended moat for incumbents.

On the flip side, we think letting Big Tech completely offload accountability to downstream users would leave a massive regulatory vacuum. Tudorache and Benifei are now attempting to thread the needle by placing stricter governance on large service providers while protecting everyday experimenters. How Europe resolves this fight in April will set the tone for tech policy worldwide: either the EU establishes a practical compromise for foundation models, or it risks setting standards so heavy-handed that the frontier gets delayed at its borders.