AI transparency on August 2, 2026: the obligation SMBs think was delayed already applies
The AI Act delay to 2027 does not cover transparency. AI content labeling and customer disclosure apply from August 2, 2026 to French SMBs and mid-sized firms.
On August 2, 2026, in just a few days, the AI Act transparency obligations enter into application. Many SMB leaders believe everything was pushed to 2027. It was not, and that confusion will cost those who let it settle in.
The Digital Omnibus, voted on June 16, 2026 and still pending Council adoption, delayed the heaviest obligations, those for high-risk systems, to December 2, 2027. That delay created a stubborn misunderstanding. In my conversations with leaders, many concluded the AI Act no longer concerned them for another two years. Yet the transparency obligations were not delayed.
Concretely, from August 2, 2026, three rules apply. A customer-facing chatbot or conversational agent must clearly disclose that it is a machine. AI-generated content, images, text, audio, deepfakes, must be identifiable as such. Emotion recognition or biometric categorization systems must be flagged to the people concerned. These obligations touch uses already common in SMBs, the website chatbot, generated marketing visuals, automated replies. The European AI Office, operational on August 1, 2026, oversees enforcement. The Digital Omnibus also extended certain facilities, sandboxes and lighter documentation, to companies under 750 employees, but a documentation facility never exempts anyone from the transparency obligation itself.
The real obstacle is not legal, it is organizational. Between 40 and 60% of real AI usage in SMBs escapes the IT department (shadow AI, field observation), and 80% of SMBs have no documented AI governance (nachnouchi.com). In other words, most leaders do not know where AI generates customer content inside their own company. You cannot label what you have not mapped, nor disclose a chatbot whose existence no one has formalized.
This is exactly what the usage inventory addresses, in week 1, listing every AI use, declared and invisible, before any compliance work. It feeds the security and sovereignty pillar of my engagements, where compliance is not a legal document produced after the fact but the direct consequence of an honest map of real usage.
Delaying high-risk systems to 2027 does not touch transparency. SMBs that wait until 2027 to act misread the calendar by two years, and will restart in 2027 with a lag rather than a lead. If you'd like to discuss it, get in touch.
Frequently asked questions
Were the AI Act transparency obligations delayed to 2027?
No. The delay obtained through the Digital Omnibus (voted June 16, 2026, still pending Council adoption) applies to high-risk systems, pushed to December 2, 2027. Transparency obligations enter into application on August 2, 2026 (AI Act text).
Which concrete obligations apply from August 2, 2026?
A customer-facing chatbot or conversational agent must disclose that it is a machine, AI-generated content (images, text, audio, deepfakes) must be identifiable, and emotion recognition or biometric categorization systems must be flagged to the people concerned.
Is a small company with no legal department really affected?
Yes, as soon as it uses a chatbot on its website, publishes AI-generated marketing visuals, or automates replies. These uses are already common in SMBs and fall within the transparency obligations, regardless of company size.
Doesn't the Digital Omnibus ease the burden for SMBs?
It extended certain facilities (sandboxes, lighter documentation, national support) to companies under 750 employees. But a documentation facility does not exempt anyone from the transparency obligation itself, which remains applicable.
Why do so many leaders not know where AI generates content in their company?
Because 40 to 60% of real AI usage in SMBs escapes the IT department (shadow AI, field observation at nachnouchi.com) and 80% of SMBs have no documented AI governance (nachnouchi.com). You cannot label content you do not know exists.
What is the link between regulatory transparency and mapping AI usage?
Compliance flows from an honest inventory. Until AI uses are listed, both declared and invisible, it is impossible to know which content must be labeled or which chatbots must be disclosed.
At which step of an AI project should this be addressed?
The Inventory step, in week 1, lists every AI use in the company before any compliance work. It feeds the security and sovereignty pillar, where compliance is the consequence of mapping real usage, not a document produced after the fact.
What does an SMB risk by waiting until 2027 to act?
It misreads the calendar by two years on transparency, which applies from August 2, 2026, and will discover non-compliant, undocumented uses under pressure in 2027, exactly the situation the current period allows it to avoid.