It’s been a long time coming, but on August 2nd, the EU AI Act stopped being a theoretical regulatory discussion and started the long road into enforcement.

In this stage of the act’s application, the European Commission begins enforcing a much broader part of the rules, including the consumer-facing transparency requirements in Article 50. The rules for most of the high-risk system requirements have been kicked into 2027 and some even further into 2028.

AI’s “cookie banner moment” per the FT, means businesses must label chatbots, synthetic media and AI-generated marketing, which means they’ll have to start producing evidence we can inspect. For software testing and quality engineering, the new transparency requirements create some obvious test conditions. Observability of disclosures, deception, integrity, and accessibility are included in a whole bunch of guidance under Article 50.

Most of the high-risk obligations have been moved to the right (again), so I anticipate that they’ll be mostly ignored until real enforcement starts, but it will be interesting to see Human in the Loop (HITL) get a run for its money. HITL won’t fly at most volumes I’ve seen if you have to demonstrate your “human editorial control” means more than someone was just available to monitor extruded output.

I also don’t think it’s possible to not look at enforcement outside of the context of performance indicators like the Future of Life Institute’s Summer 2026 AI Safety Index. Awkwardly, the Index found that even though the EU” is a leader in AI safety regulation, the top European AI company Mistral scored dead last on safety.”, which looks like Europe has apparently produced both the strictest rules for AI and the company least interested in reading them.

From the report “While there is good work being done on AI safety in the industry, the capabilities race has become more extreme. Companies have backed away from earlier commitments to release new systems only with safety measures appropriate for their capability levels; now, they’re planning to release them even if it’s demonstrably unsafe to do so.”

Safety commitments are being weakened, frameworks lack independent auditing and clear decision authority, and existential safety remains the weakest category across the industry. Fertile investigative ground for all those people the AP reported the EU is adding to its AI Office compliance and whistleblowing.

More to come on this, but it’s time to dig in testers and get your red pen out…