August 2 and the arrival of an AI regulator with teeth — the enforcement switch, not the rulebook, is the event

For a year the AI Act's general-purpose obligations were law that could not bite. On 2 August that changes, and the change is retroactive in the only sense that matters: the obligations were never suspended, only unenforceable. This week they acquire a regulator.

There is a species of law that exists without consequence, and it quietly teaches everyone the wrong lesson for as long as it lasts. On 2 August the European Commission gains the standing to investigate general-purpose model providers, order corrective measures, and fine them up to €15 million or 3% of worldwide turnover — one year to the day after the obligations took effect.

What a year of unenforceable law does

It sorts firms into two groups: the ones who complied because the obligation existed, and the ones who waited to see whether anyone would come. Both now face the same regulator, but they arrive with very different amounts of work behind them — and the second group has been accumulating exposure the entire time, because the rules were live even while the enforcer was not.

The reflex will be to treat 2 August as a start date. It is not. It is the date on which a year of accumulated position becomes actionable.

The split calendar is a tell

The Digital Omnibus pushed the high-risk obligations to December 2027 while pointedly leaving the general-purpose rules on the August timetable. That split is the Commission telling you which obligations it judged ready to enforce. A firm that reads "AI Act delayed" and stops its general-purpose compliance work will discover on Saturday that the finable part arrived exactly on time.

A law nobody could enforce taught the market that the deadline was the thing to watch. The lesson of the next quarter is that the obligation was.

European Commission — Commission starts enforcing AI Act rules on 2 August → · Technology.org — EU AI Act: what actually applies on 2 August 2026 →