When the calendar of the European Union's AI Act was drawn up, August 2, 2026 stood out on every compliance team's dashboard as an almost apocalyptic date. It was supposed to mark the moment when a regulatory regime built over years would finally stop being a document and start producing concrete effects on developers, companies and citizens. The date has arrived, and something has indeed begun: the transparency obligations of Article 50 became enforceable, and the AI Office, together with national authorities, gained oversight powers over general-purpose AI models, the so-called GPAI. But the day is far from the big impact the entire sector had prepared for, because the heaviest obligations — those covering high-risk systems — were pushed to 2027 and 2028 by amendments adopted earlier this year. The result is a transition phase that resembles a rehearsal more than a regulatory revolution.
What actually changes from now on deserves concrete attention. In the transparency field, Article 50 obliges providers and deployers to make it unmistakable when someone is interacting with a chatbot, when a piece of content is synthetically generated, and when an image or video has been manipulated enough to qualify as a deepfake. This applies to the conversational systems already scattered across customer service, health apps, education and personal assistants. Synthetic content labelling, in turn, becomes mandatory and is expected to let users and platforms identify the artificial origin of texts, audio and video. At the same time, the AI Office and national authorities begin to exercise enforcement powers over general-purpose models, opening the door to compliance measures, requests for documentation and, in more serious cases, sanctions.
The core point, however, is that Article 50 is only a fraction of the regulatory edifice. The costliest and hardest obligations, tied to high-risk systems, were moved to 2027 and 2028 by amendments passed in early 2026. That means products used in recruitment, credit, health, education and critical infrastructure keep operating, for now, without the full layer of requirements the original regime envisioned. For companies, the reading is twofold. On one side there is financial and engineering relief, because the delay offers more time to adapt conformity-assessment processes and technical documentation. On the other, there is the risk of complacency: postponing obligations can lead organizations to treat this new phase as something minor and postpone until the end a growing pile of accumulated requirements.
There is also a strategic dimension that companies must weigh. Transparency obligations are not merely formal. Chatbots that hide their artificial nature, deepfakes circulating without labels and unidentified synthetic content feed exactly the disinformation problems European lawmakers meant to contain. The way companies implement these requirements now, voluntarily and carefully, tends to set the benchmark for what will be demanded with far more force once the high-risk phase finally takes effect. Firms that treat transparency as a bureaucratic detail will probably reap regulatory trouble later, while those that see it as a competitive value may turn compliance into a trust advantage with consumers and partners.
The question that remains open is whether the calendar will hold. Postponing the high-risk phase was a political decision that answered industry pressure and fears about European competitiveness against rivals such as the United States and China. If implementation of the transparency phase generates practical resistance, operational confusion or accusations of regulatory overreach, it is plausible that new amendments will appear before 2027 to dilute the regime even further. On the other hand, if GPAI enforcement and synthetic content labelling work smoothly, the Commission could gain political capital to tighten its tone. August 2 is therefore less a dividing line than a signpost: it indicates the direction of the journey, but it does not reveal how much time, effort and negotiation are still needed before the European AI Act becomes, in practice, a full force over the market.
Sources: AIActo, NetNordic, Digital Applied
✓ Independent sources cross-checked and verified before publishing
