Regulatory · 11 Jul 2026
The AI Act dates after the Digital Omnibus
The window moved later than it first looked, and incumbents will use it.
Specifics
The numbers, and where they come from
| Quantity | Value | Basis |
|---|---|---|
| Annex III stand-alone high-risk | 2 Dec 2027 | Cited |
| Annex I embedded high-risk | 2 Aug 2028 | Cited |
| Article 50 transparency | 2 Aug 2026 | Cited |
| Article 50(2) marking | 2 Dec 2026 | Cited |
The dates
What moved and what did not
- 01
Annex III stand-alone high-risk, 2 December 2027
Deferred. This is the date most regulated conversational deployments are working towards, and it is the one our own deployment target is aligned to.
- 02
Annex I embedded high-risk, 2 August 2028
Deferred further. Relevant to anyone putting a conversational engine inside a regulated product rather than beside one.
- 03
Article 50 transparency, 2 August 2026
Unmoved. Disclosure that a user is interacting with a machine, before the first exchange.
- 04
Article 50(2) and new prohibitions, 2 December 2026
Unmoved. Machine-readable marking of synthetic content.
Our position
Stated against our own interest
Article 50 applies to us, and determinism buys nothing there. A deterministic engine must disclose that it is a machine on exactly the same terms as a transformer. We have seen this claimed as a differentiator and it is not one.
The deferral is not good news for a company at our stage. A later window favours incumbents with deployments already running, who now have longer to retrofit record-keeping. Our advantage is that Article 12 records fall out of the stage eight log natively rather than being bolted on, and a longer runway erodes that advantage rather than protecting it.
Not legal advice
This page is a dated reading of the regulation by the company it applies to. It is reviewed on a stated cadence and the review date is shown. Do not rely on it in place of counsel.