The European Union's AI Act was supposed to be the gold standard. The regulatory framework that proved democracies could govern artificial intelligence without killing innovation. It took years of negotiation, hundreds of amendments, and enough lobbying meetings to fill a stadium. And now, before the rules even take effect, companies are finding the emergency exits.

Here is the problem: the European Parliament and Council have proposed pushing back key provisions from August 2026 to as late as August 2028. That alone would be concerning. But combined with Article 111, which makes the rules non-retroactive, it creates something far worse: a permanent escape hatch.

Any AI system deployed before the new deadlines does not have to comply. Ever. Unless it is "substantially modified." A hiring algorithm that screens out candidates based on biased data? If it went live before December 2027, it could legally operate outside the AI Act indefinitely. MEP Sergey Lagodinsky called this provision "a loophole" and "a weak spot" in the law. That is diplomatic language for a structural failure.

The incentive structure is now backwards. Companies have every reason to rush their riskiest AI systems to market before the deadline. Not after. Bram Vranken from Corporate Europe Observatory put it plainly: "Some companies might abuse this timeline and quickly push risky AI systems onto the market without having to comply with the Act." The very regulation designed to prevent harm is now incentivizing a race to deploy before oversight arrives.

It gets worse. New analysis from Corporate Europe Observatory and LobbyControl found that 69% of Commission meetings on AI in 2025 were with business groups. Only 16% were with NGOs. The people writing the delays are the same people who benefit from them.

Laura Caroli, who co-negotiated the original AI Act, offered a chilling example: "A chatbot in a doll could tell a child to do something harmful, and nobody would be able to hold the manufacturer accountable" until the underlying safety law catches up.

This is the pattern with AI regulation everywhere. The United States cannot agree on a federal framework. Individual states are writing contradictory rules. And now Europe, which was supposed to lead, is giving companies a two-year head start to deploy before the rules apply. The AI Act is not dead. But it is arriving late to a party that started without it, and the damage may already be done by the time it walks through the door.

Source: TechPolicy.Press investigation with MEP Lagodinsky, Corporate Europe Observatory, and former AI Act negotiator Laura Caroli.