Anoth­er import­ant mile­stone in the imple­ment­a­tion of the AI Act is approach­ing: on 2 August 2026, the trans­par­ency oblig­a­tions set out in Art­icle 50 AI Act will become applic­able.

This marks the next stage in the phased entry into applic­a­tion of the AI Act. Earli­er phases already intro­duced the rules on pro­hib­ited AI sys­tems, AI lit­er­acy and gen­er­al-pur­pose AI, as dis­cussed in our pre­vi­ous blog posts (link and link). In this blog, we explain what the new trans­par­ency oblig­a­tions mean for organ­iz­a­tions.

 

Which AI sys­tems are sub­ject to the trans­par­ency oblig­a­tions?

Not every AI sys­tem falls with­in the scope of the trans­par­ency oblig­a­tions under the AI Act. The AI Act dis­tin­guishes between dif­fer­ent cat­egor­ies of AI sys­tems, includ­ing pro­hib­ited AI sys­tems, high-risk AI sys­tems, AI sys­tems sub­ject to trans­par­ency oblig­a­tions, and AI sys­tems present­ing lim­ited or min­im­al risk. The trans­par­ency oblig­a­tions apply exclus­ively to the third cat­egory.

It is import­ant to note that an AI sys­tem sub­ject to a trans­par­ency oblig­a­tion may also fall with­in anoth­er risk cat­egory, such as the cat­egory of high-risk AI sys­tems. Organ­iz­a­tions should there­fore care­fully assess which oblig­a­tions apply to a spe­cif­ic AI sys­tem.

 

What do the trans­par­ency oblig­a­tions require?

The trans­par­ency oblig­a­tions are inten­ded to ensure that indi­vidu­als can under­stand when they are inter­act­ing with AI and when con­tent has been gen­er­ated or manip­u­lated by AI. These oblig­a­tions apply to four types of AI sys­tems:

1. AI sys­tems that inter­act dir­ectly with indi­vidu­als;
2. AI sys­tems that gen­er­ate syn­thet­ic con­tent;
3. AI sys­tems used for emo­tion recog­ni­tion or bio­met­ric cat­egor­iz­a­tion; and
4. AI sys­tems used to cre­ate deep­fakes or gen­er­ate text on mat­ters of pub­lic interest.

The trans­par­ency oblig­a­tions under the AI Act are divided between pro­viders of AI sys­tems and organ­isa­tions that use those sys­tems (deploy­ers). Broadly speak­ing, pro­viders are respons­ible for incor­por­at­ing the tech­nic­al fea­tures and safe­guards that enable trans­par­ency, while deploy­ers are respons­ible for ensur­ing that the required trans­par­ency inform­a­tion is com­mu­nic­ated to affected indi­vidu­als in prac­tice.

 

Prac­tic­al guid­ance for com­pli­ance with the trans­par­ency oblig­a­tions

To assist organ­iz­a­tions in com­ply­ing with the trans­par­ency oblig­a­tions, the European Com­mis­sion has pub­lished Guidelines on the application of the AI Act transparency obligations. While these guidelines are not leg­ally bind­ing, they provide use­ful prac­tic­al guid­ance on com­pli­ance. The European Com­mis­sion explains, among oth­er things, that:

• indi­vidu­als should be informed that they are inter­act­ing with AI no later than the first inter­ac­tion with the AI sys­tem. A gen­er­ic ref­er­ence to AI use in terms and con­di­tions or on a web­site will gen­er­ally not be suf­fi­cient;
• inform­a­tion regard­ing the use of AI should be presen­ted in a clear, vis­ible and under­stand­able man­ner, for example through labels, ban­ners or badges that remain vis­ible through­out the inter­ac­tion with the AI sys­tem. The European Com­mis­sion has also pub­lished a set of EU icons for labelling AI-gen­er­ated con­tent that organ­iz­a­tions may use; and
• organ­iz­a­tions may choose to adhere to a vol­un­tary code of con­duct, such as the Code of Practice on Transparency of AI-Generated Content. Par­ti­cip­a­tion in such ini­ti­at­ives may help demon­strate com­pli­ance with the trans­par­ency oblig­a­tions.

 

Trans­ition­al regime under the Digit­al Omni­bus on AI

On 27 July 2026, the Digital Omnibus on AI was offi­cially pub­lished in the Offi­cial Journ­al of the European Uni­on. As a res­ult, a lim­ited trans­ition­al regime has entered into force for the trans­par­ency oblig­a­tion con­tained in Art­icle 50(2) AI Act.

Under this trans­ition­al arrange­ment, AI sys­tems that are sub­ject to this oblig­a­tion and that were placed on the mar­ket before 2 August 2026 bene­fit from a four-month trans­ition peri­od. This provides pro­viders with addi­tion­al time to bring their sys­tems into com­pli­ance with the new require­ments.

 

What should all organ­iz­a­tions do?

The trans­par­ency oblig­a­tions under the AI Act do not only affect developers of AI sys­tems. They may also apply to organ­iz­a­tions that pro­cure and use AI sys­tems. Although this may seem obvi­ous, it is essen­tial that organ­iz­a­tions have a clear over­view of the AI sys­tems used with­in their organ­iz­a­tion. Such an invent­ory is not only an import­ant com­pon­ent of an AI gov­ernance and risk man­age­ment frame­work, but is also neces­sary to determ­ine wheth­er trans­par­ency oblig­a­tions apply.

 

AI Code of Con­duct

In addi­tion, organ­iz­a­tions of all sizes should estab­lish clear intern­al rules gov­ern­ing the devel­op­ment and use of AI. A prac­tic­al way to achieve this is by imple­ment­ing an AI Code of Con­duct. Such a policy doc­u­ment can explain what AI is, the risks asso­ci­ated with its use, and the rules that apply with­in the organ­iz­a­tion for the respons­ible use of AI. For employ­ees involved in the devel­op­ment or imple­ment­a­tion of AI sys­tems, an AI Code of Con­duct can also help raise aware­ness of the trans­par­ency oblig­a­tions under the AI Act. More broadly, it sup­ports AI lit­er­acy among employ­ees, which has already been a require­ment under the AI Act since 2 Feb­ru­ary 2025.

Do you have ques­tions about the upcom­ing 2 August dead­line, or would you like assist­ance with draft­ing an AI Code of Con­duct? HVG Law would be pleased to assist. Please feel free to con­tact us to dis­cuss how we can sup­port your organ­iz­a­tion.