Artificial intelligence is no longer just a matter of operational efficiency: it has become a matter of governance and risk management. 2 August 2026 marks a significant milestone for the European market, because Article 50 of the AI Act (EU Regulation 2024/1689) becomes applicable, introducing specific transparency obligations for AI systems intended to interact with people or to generate content. In Italy the picture is completed by Law 132/2025, which since 10 October 2025 has already required disclosure of the use of AI in professional services.
Unlike the prohibitions that took effect at the end of 2024, this deadline touches the day-to-day operations of many businesses – SMEs included – that use AI tools in their dealings with customers and suppliers. If your company employs systems of this kind, now is the time to check where you stand.
Who the new obligations apply to
The rules distinguish two roles, and it is a distinction that concretely changes what you have to do. The provider (whoever develops or places the system on the market) carries the more technical obligations, in particular marking artificially generated content in a machine-readable format. The deployer (whoever uses the system in their own business) has more limited obligations: clearly informing people when they are interacting with an AI system, and when a piece of content has been artificially generated or manipulated. Most businesses that adopt off-the-shelf solutions fall into the second group. Confusing the two levels leads companies to fear obligations that, in practice, do not fall on the user.
In concrete terms, the transparency obligations concern systems that interact with people (chatbots, virtual assistants for support, sales or first contact), where the user must know they are dealing with an automated system; AI-generated content (text, images, audio, video) meant to inform the public, which must be made recognisable as artificial, except for clearly creative, satirical or editorial uses subject to specific rules; and biometric recognition or emotion-analysis systems, where permitted, which require disclosure to the people involved. The underlying principle is simple: anyone interacting with a machine, or reading content produced by a machine, has the right to know it.
An essential check for those who run the business
Three points deserve an immediate review. First, clear and timely disclosure: clauses buried in the terms of use are not enough – the notice must be explicit and present at the start of the interaction. Second, recognisability of content: AI-generated content used for informational or commercial purposes must be identifiable as such, and the technical marking falls above all on the system’s provider. Third, internal mapping and skills: it is worth taking stock of the AI systems in use and ensuring staff have a minimum level of training on the limits and risks (so-called AI literacy).
A concrete operational reference already exists: the Code of Practice on Transparency of AI-Generated Content, published by the European Commission on 10 June 2026. Adherence is voluntary, but the obligations of Article 50 remain binding regardless. The practical value of the code is defined by the Regulation itself: once the Commission and the European AI Board have confirmed its adequacy – the assessment is under way – signatories will be able to rely on the code to demonstrate compliance across the Union, reducing uncertainty about what the authorities expect. One useful calendar detail: organisations wishing to appear on the first public list of signatories must submit the form to the Commission by 22 July 2026; signing remains possible afterwards. The Commission has also released a set of icons for labelling AI-generated content and will publish, before 2 August, guidelines on the scope of the obligations.
A note on the timeline (useful, given the confusion of recent weeks)
You may have read that the AI Act was “postponed”. That is an oversimplification. In November 2025 the Commission proposed the so-called Digital Omnibus; the European Parliament approved the text on 16 June 2026, the Council gave its green light at the end of June, and publication in the Official Journal of the EU is expected in the course of July, in time for the August deadline. The measure defers the application of the rules on high-risk systems (to 2 December 2027 for standalone systems, to 2 August 2028 for those embedded in regulated products). The transparency obligations of Article 50, however, remain fixed at 2 August 2026: the deferral does not touch them. The only relief provided concerns content-generation systems already on the market before August, for which marking is due from 2 December 2026. In short: those who headline that “the AI Act has been postponed” are looking at the high-risk part; the part concerning transparency towards the public is operational from the scheduled date. And a rule of prudence applies: until the amendment is published in the Official Journal, the law in force remains the original text.
The deferral is no exemption from governance
There is also a reason – rooted in Italian law – not to read the deferral as permission to wait. Under Italian company law, directors are liable to the company for the diligence with which they manage it (Article 2392 of the Civil Code), a principle that long predates the AI Act. Adopting artificial intelligence systems without oversight – without knowing which tools are in use, what data they process, what risks they introduce – is a management choice, and it is judged as such. The European calendar sets when inspections and penalties arrive; the duty to govern the company’s risks already exists, and nobody has postponed it.
The penalty regime, without alarmism
The highest thresholds set by the AI Act – up to €35 million or 7% of total worldwide annual turnover – concern prohibited practices (Article 5), not the transparency obligations. Breaching the transparency obligations falls under a separate regime (Article 99(4)): up to €15 million or 3% of turnover. And for SMEs and start-ups the Regulation expressly provides that the lower of the fixed amount and the percentage applies – relief designed precisely for smaller businesses. The fact remains that non-compliance still exposes you to penalties and, above all, affects the value of the business in the contexts that matter.
Compliance costs need not fall entirely on the company
For most SMEs, complying with the transparency obligations has a defined scope: mapping the systems in use, fixing the disclosures, training people. The costs are contained, and part of them need not weigh on the income statement: AI training qualifies for funding through Italy’s joint interprofessional funds, and subsidised-finance measures for digitalisation – national and regional – in many cases also cover compliance work. Identifying the right measure for one’s situation is a job in itself, and it is often what turns an obligation into an opportunity to put things in order.
The perspective of the district and of international relations
For businesses in the Prato district and for those operating on Italy-abroad relationships – including the Chinese-Italian business community – compliance with the AI Act is not merely a regulatory matter. In a sale, merger or entry of new shareholders, the soundness of technological compliance becomes part of the due diligence. A company that is not in order does not only risk penalties: it loses ground when its value is examined by a counterparty.
GCompass’s approach
Getting into compliance does not mean holding innovation back, but securing it to safeguard business continuity. Reducing exposure to regulatory risk is the first step towards building orderly growth. If you want to understand how the new AI Act obligations apply to your company and what corrective measures to adopt before 2 August 2026 – and which subsidised-finance instruments can cover the costs – let’s talk – no commitment.
GCompass & Partners – Strategic Corporate Advisory. Management consulting pursuant to Law 4/2013. → gcompassrl.info/en/contatti/
Frequently asked questions
When do the AI Act transparency obligations start to apply?
On 2 August 2026: from that date Article 50 of EU Regulation 2024/1689 applies, requiring people to be informed when they interact with an AI system and AI-generated content to be made recognisable. In Italy, the duty to disclose the use of AI in professional services has been in force since 10 October 2025 (Law 132/2025).
Has the AI Act been postponed?
Only in part: the Digital Omnibus defers the obligations on high-risk systems (to 2 December 2027 for stand-alone systems, to 2 August 2028 for those embedded in products), while the transparency obligations of Article 50 remain set for 2 August 2026. The European Parliament approved the text on 16 June 2026 and publication in the Official Journal is expected in July; until then, the original Regulation remains in force.
My business only uses off-the-shelf AI tools: does it still have obligations?
Yes: whoever uses AI systems in their own business (the deployer) must clearly inform the people interacting with the system and disclose artificially generated or manipulated content. The heavier technical obligations, such as machine-readable marking, fall on whoever develops or sells the system.
What penalties does an SME risk for breaching the transparency obligations?
Up to EUR 15 million or 3% of total worldwide annual turnover (Article 99(4) of the AI Act); for SMEs and start-ups, the lower of the two expressly applies. The higher thresholds often quoted – EUR 35 million or 7% – concern the prohibited practices of Article 5, not transparency.
What is the Code of Practice on Transparency, and is it worth signing?
It is the voluntary code published by the European Commission on 10 June 2026 to implement the Article 50 transparency obligations in practice. Once its adequacy is confirmed by the Commission and the European AI Board (assessment under way), signatories will be able to rely on the code to demonstrate compliance; organisations wishing to appear on the first public list of signatories must submit the form by 22 July 2026.
Who is accountable within the company if AI systems are not compliant?
The directors, under the general duty of diligent management (Article 2392 of the Italian Civil Code): adopting AI systems without knowing which are in use, with what data and with what risks is a management choice, and it is judged as such – regardless of the European timetable.
Can the costs of AI Act compliance be financed?
Partly, yes: AI training qualifies for funding through Italy’s joint interprofessional funds, and subsidised-finance measures for digitalisation – national and regional – in many cases also cover compliance work. The right measure depends on the individual company’s situation.
The practical guide to the AI Act for your business
Download our free guide: what changes on 2 August 2026, the operational checklist, the real penalties for SMEs and the five myths to debunk – with concrete SME examples.