The AI Act's August 2026 deadline
Transparency duties now apply, while the timetable for high risk systems has moved. For software companies, the two changes require different responses.
On 2 August 2026, important parts of the EU AI Act began applying, including transparency obligations for certain AI systems. The European Commission's AI Office and national authorities also began enforcing the Act under the new framework. For a company selling software in Europe, this made the way an AI feature presents itself to users an immediate product question.
An amendment known as the AI Omnibus had entered into force on 27 July, extending the timetable for high risk systems. Companies faced immediate disclosure obligations alongside a longer preparation period for high risk requirements. Each part of the Act needed its own compliance plan.
Two different timetables
The Commission's announcement established the new sequence. Rules for high risk uses listed in Annex III, including certain uses in employment, education and critical infrastructure, are due to apply from 2 December 2027. Rules for high risk AI embedded in regulated products covered by Annex I are due from 2 August 2028. Those extensions give developers more time to prepare for the relevant requirements and supporting standards.
- 27 July 2026
AI Omnibus enters into force
The amendment changes the timetable and several administrative requirements.
- 2 August 2026
Transparency obligations apply
Article 50 covers specified AI interactions and generated or manipulated content.
- 2 December 2026
Marking transition ends
Providers of generative systems placed on the market before 2 August must comply with Article 50(2).
- 2 December 2027
Annex III high risk rules
The revised date for high risk uses in the areas listed in Annex III.
- 2 August 2028
Annex I product rules
The revised date for high risk AI embedded in the regulated products covered by Annex I.
Application dates published by the European Commission. The later dates concern defined categories of high risk systems, not a general suspension of the Act.
The transparency obligations already in force concern specified interactions and types of content. Their scope depends on how a system is used. The Commission also distinguishes these system obligations from the rules for general purpose AI models, which began applying in August 2025. A model supplier, an application developer and the organisation using an application may consequently have different responsibilities.
There is a narrower transition for machine readable marking. Regulation (EU) 2026/1744 gives providers of generative systems placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). That extension concerns the marking of synthetic content by those existing systems. It does not postpone the other Article 50 transparency duties.
Where transparency enters the product
The Commission's guidance identifies obligations for both providers and deployers. Providers of systems that interact directly with people must ensure that individuals are informed when they are interacting with AI, within the scope and exceptions described by the rules. Providers of relevant generative systems must support detection through machine readable marking of generated or manipulated content. Deployers have disclosure duties for specified uses, including deepfakes and certain public interest text.
These are different engineering tasks. A notice in a customer service conversation concerns the user's understanding of the interaction. A machine readable mark concerns what downstream software can detect in an output. A publication disclosure concerns how an audience encounters content. A single notice in general terms and conditions cannot be assumed to perform all three functions.
The exceptions matter as much as the headline requirements. The Commission's guidance discusses standard editing and distinguishes public interest text without human review or editorial control. Product teams need a record of how their particular feature is classified and used. Labelling every output indiscriminately can obscure the distinctions users need to understand; omitting a required disclosure can create a more direct problem.
A supplier contract cannot describe every use
Enterprise AI often reaches the customer through several businesses. One supplies a model, another adds retrieval or logic for a particular task, and another incorporates the result into a workflow. The Commission's guidance addresses responsibilities along this chain. An assessment starts with who provides the system, who deploys it and what each party controls.
For investors, a useful diligence exercise is to follow one output from generation to delivery. Establish where marking is applied, what happens when content is exported, and which party presents any necessary disclosure. Review an actual customer deployment as well as the developer's demonstration environment. A product can behave differently when a customer changes the interface, downloads an image or distributes text through another application.
Commercial agreements should make those dependencies understandable. A buyer needs to know which controls arrive with the product and which must be implemented in its own processes. Version changes also matter: replacing a model or adding an output format can change the assumptions behind an earlier assessment. The resulting support and integration work belongs in the cost of serving the customer.
What the extension changes for investment
The revised high risk timetable changes the timing of expenditure. It does not establish that a proposed use will be commercially viable or easy to approve. A company developing an employment tool or a safety component still needs a credible understanding of its intended use, customer responsibilities and evidence requirements. Extra preparation time is valuable when it is attached to a funded work programme.
The amendment also extends some measures previously reserved for smaller businesses to small midcap companies and expands opportunities for supervised testing. Those provisions may help companies prepare. Investors still need evidence of customer demand, the work required for deployment and the cost of ongoing oversight, even when a company has access to a regulatory sandbox.
Compliance with the transparency requirements should be visible in the product and its operating records. Users receive the information relevant to their interaction. Content retains the marking or disclosure required for its use. Customers understand their responsibilities. These details allow investors to assess a company's readiness against its stated compliance position.
Investment implications
- Article 50 transparency obligations now apply. Certain existing generative systems have until 2 December 2026 to meet the machine readable marking requirement.
- The AI Omnibus moved the main Annex III high risk date to 2 December 2027 and the Annex I product date to 2 August 2028.
- Product diligence should examine actual interfaces, content handling and supplier responsibilities, alongside the legal assessment.
Sources
- AI Omnibus enters into force (opens in a new tab)
European Commission
- Commission starts enforcing AI Act rules and new transparency requirements on 2 August (opens in a new tab)
European Commission
- Guidelines on transparency obligations for providers and deployers of certain AI systems (opens in a new tab)
European Commission
- AI Act: application timeline and regulatory framework (opens in a new tab)
European Commission
- Regulation (EU) 2026/1744 (Digital Omnibus on AI) (opens in a new tab)
European Union
Published
