On 2 August 2026 the European Commission’s AI Office and the national competent authorities began enforcing the AI Act, and the new transparency rules took effect. After two years of drafting, consultation and postponement, the operative date has passed.
What users see
Chatbots must disclose that they are AI. Deepfakes must be labelled. AI-generated or altered content must carry machine-readable marks. More than 180 organisations have signed the Code of Practice giving effect to these transparency rules.
Three ways in
The AI Office launched three reporting channels at the same time: a Complaint Tool for reported infringements by providers and deployers under Article 85; a dedicated channel for downstream providers, covering Articles 53 to 55 and Article 89(2); and a Whistleblower Tool offering secure reporting for workers.
The third is the one to watch. Regulatory regimes of this kind are rarely activated by inspectors, who lack the access and the numbers. They are activated by insiders. A secure channel for employees does more practical enforcement work than a headline fine, and it changes the internal incentives of every organisation that deploys these systems.
For model providers
Providers of general-purpose AI models are now subject to documentation requirements, a copyright policy obligation, and a published summary of training data. Models presenting systemic risk carry further obligations. Fines run to 3% of global annual turnover.
Housekeeping, and a signal
The AI Office opened a call for around 40 contract-agent posts, with applications closing on 8 September 2026 – simultaneously an admission about capacity and an opening for lawyers with genuine technical training. Under the AI Omnibus, obligations for high-risk systems remain postponed to 2 December 2027, and to August 2028 for high-risk systems integrated into regulated products. Roberto Viola, who has directed DG CNECT since 2015, has had his retirement extended to 30 September 2027.
What this means outside the Union
For a Georgian provider or faculty the practical position has changed in one respect. Until 2 August the AI Act was a text to prepare for. It is now a regime with live channels through which a complaint can be filed against a provider whose output reaches the Union – by a competitor, a downstream integrator, or an employee. The sensible response is unglamorous: know which of your systems generate content, know whether they are marked, and be able to produce documentation without a scramble.
Drawn from The EU AI Act Newsletter #108, 12 August 2026.