Your team may soon paste text from ChatGPT into a customer proposal that carries an invisible signal of AI involvement.
That signal will not tell you whether the proposal is accurate, who reviewed it or whether it should have been sent. Equally, the absence of a detected watermark does not prove that a human wrote it.
For Irish business leaders, the announcement is a prompt to review how AI-generated content is used, checked and approved.
What has OpenAI announced?
On 5 October, OpenAI announced a phased rollout of invisible watermarks for eligible ChatGPT and Codex text outputs in the EU over the coming weeks.
Customers worldwide using OpenAI’s API – the connection that allows software to access its models – can opt in for selected models. Watermarking remains off by default for API use.
Access to the text detector will initially be limited to approved researchers and expert organisations. Businesses should not expect a public tool for checking documents at launch.
How does text watermarking work?
OpenAI’s system, textGrain, introduces a statistical pattern into the model’s word choices. A detector looks for that pattern.
The watermark does not identify the user, establish responsibility or verify accuracy. Detection can also miss short, edited or translated passages, and false positives are possible.
A detected watermark is a limited signal of AI involvement. A missing watermark is not proof of human authorship.
What does the EU AI Act require?
The EU AI Act‘s Article 50 transparency obligations have applied since 2 August 2026. A limited grace period runs until 2 December 2026 for the marking and detection obligation for systems placed on the market before 2 August.
The distinction between machine-readable marking and disclosure to an audience matters. Providers have obligations to make generated content identifiable. Separate rules govern when people must be informed about AI interactions or synthetic content.
Article 50 does not require every AI-assisted email, tender or proposal to carry a label. For text, the publication disclosure duty concerns content intended to inform the public on matters of public interest. An exception applies where the content undergoes human review or editorial control and a person or organisation holds editorial responsibility.
That review must be substantive: checking spelling or grammar alone is insufficient. Where disclosure is required, an invisible watermark does not replace a clear notice to the audience.
Two mistakes that leaders should avoid
1. Treating a watermark as approval
A sales proposal still needs someone to check its pricing, delivery commitments and customer claims. A watermark cannot perform that review.
Assign a named person responsibility for checking and approving the final document.
2. Making detection your entire AI policy
Your organisation needs a way to record AI use without relying on a detector to identify it afterwards.
For work where errors could have significant consequences, keep a proportionate record of:
- The AI tool used.
- The content or task it supported.
- Who reviewed and approved the final output.
Build this into existing approval processes so staff can follow it without unnecessary administration.
What to do this week
Ask your technology and compliance teams to prepare a one-page overview covering:
- Which teams use ChatGPT, Codex or other AI writing tools.
- Where generated text enters customer communications, public content or regulated workflows.
- Who checks and approves those outputs.
- Whether eligible OpenAI API integrations should have watermarking enabled.
- Which workflows require disclosure and how it is provided.
Then add a clear rule to your acceptable-use policy:
Where AI disclosure is required, provide it clearly to the intended audience. An invisible watermark does not replace disclosure or human review.
The question for your leadership team is practical: Can we explain where AI contributed, who checked the work and who approved it?
Turn AI policy into everyday practice
AI Ireland helps leadership teams put AI adoption and governance into practice. If your organisation needs clearer policies, better staff understanding or a review of its AI workflows, contact AI Ireland to discuss a tailored leadership session.
Discover more from AI Ireland
Subscribe to get the latest posts sent to your email.
