Transparency is the watchword.
For European companies, the AI Act is entering its most operational phase. The new provisions affect not only those who develop AI models and systems, but also companies that integrate them into their services, use them in interactions with customers and employees, or produce content for the public.
Starting from August 2, 2026, another piece will be added to the AI Act , already in force since August 1, 2024. The first bans, for systems with ” unacceptable risk ” (mass surveillance, social scoring systems, predictive policing and the like) have already come into force since February 2, 2025. From August 2, 2026, however, most of the provisions will come into force, starting with the transparency obligations towards users.
AI Act: Transparency requirements on AI use begin August 2nd.
August 2nd will be a key date for European AI regulations. Companies will be required to be more transparent about their use of AI. This applies to various types of services, such as chatbots and virtual assistants, for example, even in call centers: users will need to be immediately informed if they are communicating with an AI rather than a real person.
But even those who produce content, as in the case of Edge9: AI can obviously be used as a support and, if desired, even to have an LLM generate entire pieces of content, whether images or text. In this case, however, it will be necessary to explicitly warn the user that the content is generated by unsupervised machines. If, however, AI is used only as a support, for example, to correct typos, for research, or in any case the piece will be reviewed by humans before being published, it will not be necessary to label it as AI-generated.
This is especially true for deep fakes : the law does not prohibit the use of AI to generate photomontages, videos, or other images, even if they depict famous people (politicians, athletes, actors). The important thing, however, is that it is clearly indicated that the images are machine-generated. This is key, because while it’s true that some content is clearly fake (think of political communication, even in Italy, where deep fakes are now the norm), in other cases it’s easy to spread misinformation by generating credible images, even if they’re fake.
What changes for those who create AI models
August 2 will also bring new rules for those distributing general -purpose AI (GPAI ) models, i.e., those not designed for a specific function. This category includes solutions such as ChatGPT, Gemini, Claude, and so on. Specifically, starting from this date, the European AI Office will have more powers. The AI Office assumed supervisory powers over these models as of August 2, 2025. A year later, the difference is that we are moving from the current dialogue phase, in which the AI Office could request clarifications and communicate with manufacturers, to the operational phase: in practice, the AI Office will have the effective power to investigate, impose corrective measures, and possibly impose sanctions on those who do not comply with the rules. Please note, however, that these rules only apply to GPAI models released on the market after August 2, 2025. Those released before that date are excluded.
While until now, companies like OpenAI, Google, and Anthopic could simply declare compliance, starting August 2, 2026, these companies must be able to provide verifiable evidence of their security and governance processes. If requested, they will also be required to provide the AI Office with a technical evaluation of the model.
The next steps: what will happen between now and 2028?
The next major step will come in 2027. GPAI models launched on the market before August 2025 will also have to comply by August 2, 2027. New requirements will apply from December 2 for many AI systems classified as high-risk, such as those used in personnel selection, access to credit, education, and essential services. Companies using AI in these areas will therefore have to introduce more rigorous procedures for risk management, data quality, human oversight, logging, and system monitoring.
In 2028, the rules will also extend to AI systems embedded in products for regulated sectors, such as medical devices and other equipment subject to conformity assessment.
L’articolo European AI Act: new obligations begin August 2nd. sembra essere il primo su 7Grammilavoro.
