The European Commission released detailed implementation guidelines on 20 July 2026 for transparency obligations under Article 50 of the EU AI Act. The document addresses providers and deployers of directly interactive AI systems, synthetic content generators, and deepfake tools, specifying how enterprises must inform users and label outputs.
Scope and Compliance Requirements in the Guidelines
The guidelines define key terms including directly interactive AI systems and AI-generated text on matters of public interest. They outline exemptions for narrow business-to-business industrial applications and standard editing tools. Enterprises must now assess whether their systems fall inside or outside these boundaries before the obligations take effect.
Practical measures include adding clear disclosures during user interactions and embedding machine-readable markers on generative outputs. For systems already on the market before 2 August 2026, a grace period extends the marking requirement until 2 December 2026. The Commission also confirmed that the voluntary Code of Practice on Transparency of AI-Generated Content, finalised in June 2026, meets the necessary standards.
National market surveillance authorities will enforce these rules, while the AI Office handles general-purpose AI model obligations. Maximum fines reach €15 million or 3 percent of global turnover, with proportional reductions available for small and medium-sized companies.
Timeline Context and Enterprise Preparation Steps
The EU AI Act entered into force on 1 August 2024. Prohibitions and AI literacy requirements applied from February 2025, followed by general-purpose AI model rules in August 2025. The Digital Omnibus Regulation, which entered into force on 27 July 2026, postponed several high-risk obligations: stand-alone systems listed in Annex III now face deadlines in December 2027, while embedded systems in Annex I shift to August 2028.
The first harmonised standard, EN 18286:2026 on quality management systems under Article 17, was cited in the Official Journal around 14 October 2026. Companies can use this standard to demonstrate conformity on a voluntary basis. Additional guidance on high-risk classification, serious incident reporting, and fundamental rights impact assessments is expected before the 2027–2028 deadlines.
Enterprises should begin by inventorying all AI systems in use, mapping provider and deployer roles, and documenting existing transparency practices. Early preparation reduces the risk of retroactive changes once high-risk rules activate. The guidelines also signal continued alignment between the AI Act and other digital regulations, helping organisations avoid overlapping compliance efforts.
Frequently asked questions
When do the Article 50 transparency obligations under the EU AI Act start to apply?
Most obligations apply from 2 August 2026. Systems placed on the market before that date receive a grace period until 2 December 2026 for machine-readable marking requirements.
Which AI systems are exempt from the new transparency guidelines?
Narrow business-to-business industrial systems and tools limited to standard editing functions fall outside the scope. The guidelines provide detailed criteria to determine eligibility.
How does the Digital Omnibus Regulation affect high-risk AI deadlines?
It delays stand-alone high-risk systems in Annex III to 2 December 2027 and embedded systems in Annex I to 2 August 2028. The changes also extend certain relief measures to small mid-cap companies.
What enforcement bodies handle Article 50 compliance?
National market surveillance authorities primarily oversee Article 50. The AI Office manages general-purpose AI model rules, and fines can reach up to 3 percent of global turnover.
By YasTech Editorial
Comments 5