Brussels: The European Union is approaching a date that will test whether its landmark Artificial Intelligence Act is a credible law or an elaborate intention. From 2 August 2026 the Commission’s AI Office gains the power to act against providers of general-purpose AI models, issuing requests for information, demanding access to systems and ultimately imposing fines of up to 3 percent of global annual turnover or 15 million euro, whichever is larger. The obligations themselves, on transparency, copyright and the management of systemic risk, took effect for new models a year earlier, but enforcement is what converts a rulebook into a constraint. That moment is now weeks away.

The instrument meant to make compliance manageable is the General-Purpose AI Code of Practice, finalised by the AI Office in 2025 and signed by roughly two dozen developers, including Anthropic, Google, Microsoft, OpenAI, Mistral, IBM and Cohere. The signatures matter less for what they promise than for what they reveal. Meta declined to sign, and Elon Musk’s xAI endorsed only the safety chapter, leaving the largest open-weight developer and one of the most prominent frontier labs partly outside the voluntary framework. A code that the biggest players can opt out of becomes, in practice, a test of whether the binding law behind it has teeth.

Here lies the contradiction at the centre of European digital policy. At the very moment enforcement begins, Brussels is trying to ease the same rules. The Digital Omnibus, on which the Parliament and Council reached a provisional agreement in May 2026, would postpone the application of the high-risk obligations, pushing standalone systems to no later than December 2027 and embedded ones to August 2028, while adding a fresh prohibition on so-called nudifier apps that generate non-consensual intimate imagery. The package is a tacit admission that the original timetable outran the readiness of both regulators and industry. Yet it is not yet adopted, which leaves companies in an uncomfortable limbo: the prudent course is to keep building toward the original August deadline that the Omnibus may or may not move.

This is the deeper tension worth naming. The Union built its reputation on being the world’s foremost digital rule-setter, the jurisdiction whose standards others quietly copy. But competitiveness anxiety, sharpened by American scale and Chinese speed, has pulled policy in the opposite direction, toward simplification and delay. The AI Act is becoming the arena where those two instincts collide. Enforce too hard and Brussels risks confirming the caricature of a continent that regulates what it cannot invent. Enforce too softly, after years of insisting the rules were necessary to protect citizens, and it forfeits the moral authority that made the Brussels Effect work in the first place.

The practical question for the coming months is what the AI Office actually does with its new powers. Credible enforcement does not require dramatic fines on day one; a few well-chosen requests for information, directed at firms that signed nothing, would signal that the code is a floor rather than a favour. Equally, a long silence would tell every developer that the August date was theatre. The Office is newly staffed and lightly resourced relative to the scale of the models it must supervise, and that mismatch is itself a policy choice that will become visible in how selectively it acts.

What makes this a defining episode rather than a procedural footnote is that the rest of the world is watching to calibrate its own approach. If Europe can show that binding AI rules and a living technology sector are compatible, it strengthens the case for regulation everywhere. If the law arrives only to be diluted before it bites, the lesson others draw will be the opposite. The coming weeks will reveal which story August 2026 tells.