OpenAI text watermarks in the EU: what changes for SMEs
In brief
OpenAI has started marking text produced by its language models, according to Silicon.fr. The function is first offered as an optional setting on the API, then is planned for ChatGPT and Codex in the European Union within a few weeks. Under the AI Act, providers must mark generated content for systems placed on the market before 2 August 2026 and have until 2 December 2026 to comply.
Source : Silicon.fr - « OpenAI bute sur le tatouage des sorties texte »
What it changes for an SME
For an SME that uses ChatGPT or the OpenAI API, watermarking is handled by the provider. The source does not describe extra steps for end users. If you publish generated text, it may be identifiable as AI-produced, though according to OpenAI detection weakens on short or partly rewritten passages. The AI Act marking duty falls on providers; it matters more if you yourself place an AI system on the market.
What you can do
- Map which OpenAI tools (ChatGPT, Codex, API) your teams use in the European Union.
- If you embed these models in a product you sell, check whether you are a provider under the AI Act and note the 2026 deadlines.
- Do not treat a watermark as definitive proof: according to OpenAI, detection weakens on short or rewritten text.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.