OpenAI to introduce invisible watermarks for EU ChatGPT users to comply with AI Act
OpenAI will begin adding invisible watermarks to text generated by ChatGPT for users in the European Union (EU) in the coming weeks. The move is part of an effort to comply with the EU AI Act, which requires AI-generated content to be identifiable. However, the detection tool will only be available to researchers and expert organizations, not the public.
Editor, Lazyfounder
30 SEC SUMMARY
- OpenAI will introduce invisible watermarks to ChatGPT-generated text for users in the European Union (EU) to comply with the EU AI Act.
- The watermarking system, called textGrain, subtly alters word choices in generated text but may reduce text quality.
- Detection tools for the watermarks will only be available to researchers and expert organizations, not the public.
- The changes are being implemented exclusively for EU users and will roll out in the coming weeks.
- Watermarking is prone to errors, especially for shorter text or after edits.
TABLE OF CONTENTS
KEY HIGHLIGHTS
- OpenAI announced on October 5 that it will introduce invisible watermarks to ChatGPT-generated text for EU users.
- The watermarking system, textGrain, alters word choices in generated text but may weaken text quality.
- Detection tools for the watermarks will be restricted to researchers and expert organizations.
- The EU AI Act requires AI-generated text to be identifiable in a machine-readable way.
- Watermarking may be less effective for shorter text or after user edits.
OpenAI’s watermarking plan for EU users
According to TechRadar, OpenAI will begin adding invisible watermarks to text generated by ChatGPT for users in the European Union (EU). The move is intended to comply with the EU AI Act, which requires AI-generated content to be identifiable in a machine-readable format. The changes are set to roll out in the coming weeks.
The watermarking system, called textGrain, works by subtly altering word choices in generated text. While the watermarks are not visible to users, OpenAI claims they can be detected by a specialized tool. However, this tool will only be available to researchers and expert organizations, not the general public.
Limitations of the watermarking system
TechRadar reports that the textGrain system is not foolproof. It is prone to errors, particularly for shorter or more constrained text. Additionally, editing the generated text—such as rephrasing or trimming—can weaken or remove the watermark entirely.
The system’s implementation may also impact text quality. According to TechRadar, the alterations made to embed the watermark could result in suboptimal output. Some observers have described the approach as a "box-ticking exercise" to meet regulatory requirements rather than a robust solution.
Why the EU AI Act requires watermarking
The EU AI Act, which came into force in August 2026, mandates that AI-generated content must be identifiable to mitigate risks such as misinformation, deepfakes, and unauthorized use. The regulation applies to all AI systems operating within the EU, regardless of where the parent company is based.
OpenAI’s decision to implement watermarking exclusively for EU users reflects the region’s stricter regulatory environment compared to markets like the US or UK. Similar requirements are not yet enforced in other major jurisdictions.
Developing: what is not yet confirmed
The following is reported but has not been independently confirmed.
TechRadar reports that some critics have described the textGrain system as an "easily broken solution full of holes." However, these claims have not been independently verified. OpenAI has not publicly addressed the specific weaknesses alleged by sources.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
OpenAI’s move to introduce watermarking for EU users is a direct response to regulatory pressure, but it raises questions about the practicality of such measures. For founders and operators, this signals that compliance with the EU AI Act may require trade-offs—such as reduced output quality or additional technical overhead—without guaranteeing foolproof detection.
The restricted access to detection tools also suggests that OpenAI is prioritizing control over transparency. This could limit the ability of journalists, educators, or businesses to verify AI-generated content unless they partner with approved experts. Meanwhile, the system’s susceptibility to errors and editing underscores that watermarking is not a silver bullet for AI accountability.
Key takeaways
- Founders building AI tools for the EU market must prepare for similar compliance requirements, including potential trade-offs in output quality or usability.
- Watermarking systems like textGrain may not be reliable for short or edited text, limiting their effectiveness for real-world use cases.
- Regulatory fragmentation is creating uneven expectations for AI developers—expect stricter rules in the EU compared to the US or UK.
- Restricting detection tools to experts could create bottlenecks for organizations that need to verify AI-generated content.
- This development highlights the tension between compliance and practicality in AI regulation.
FAQ
Will the watermarking affect ChatGPT’s performance or response quality?
According to TechRadar, the watermarking system may result in suboptimal text output due to the subtle alterations made to embed the watermark.
Can users outside the EU access the watermarked version of ChatGPT?
No. OpenAI’s changes are being implemented exclusively for users in the European Union to comply with the EU AI Act.
How can I check if a text was generated by ChatGPT?
The detection tool for identifying watermarks will only be available to researchers and expert organizations, not the public. Users cannot verify watermarks on their own.
Related on Lazyfounder
Sources
- TechRadar · 2026-10-13
ChatGPT will soon start secretly watermarking its answers — here's what that means for the text you copy and share
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.
