OpenAI Introduces Text Watermarking for ChatGPT and Codex to Meet EU AI Act Requirements
11 Sources
[1]
OpenAI will start watermarking ChatGPT's text in the EU
OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act, the company said Monday in a blog post. The EU AI Act's transparency rules, which took effect on August 2, require AI companies to mark AI-generated content in
[2]
OpenAI is adding text watermarking in ChatGPT and Codex
An invisible, machine-readable watermark in text output is rolling out to ChatGPT and Codex, but only for users in the European Union at first. OpenAI says its textGrain watermarking "matched or exceeded" other approaches like Google DeepMind's SynthID for text, which is also the basis for the
[3]
OpenAI to Add Watermarks on ChatGPT's Text Outputs, But Only in the EU
ChatGPT will start including an "invisible watermark" on text outputs to comply with European Union regulations, OpenAI announced on Monday, but it will only apply to users in the EU. "In the EU, we'll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the
[4]
OpenAI is adding invisible watermarks to ChatGPT and Codex text in the EU
OpenAI is preparing to add invisible watermarks to text generated by ChatGPT and Codex in the European Union. The watermark will not be visible when you read or copy the text. Instead, OpenAI says its new textGrain technology slightly changes the model's word choices to create a statistical
[5]
OpenAI will add a digital watermark to text and code generated in the EU - Engadget
OpenAI just announced its plans to comply with part of the European Union's new AI transparency rules. The company will soon begin adding an invisible watermark to text and code generated in the EU. This follows a similar move by Anthropic earlier this year. OpenAI is using a proprietary system
[6]
OpenAI brings text watermarking to ChatGPT and Codex in the EU
Its own tests show editing the text weakens detection. Text watermarking stays opt-in for API customers worldwide, and the detector is limited to approved researchers. OpenAI will add an invisible watermark to text from ChatGPT and Codex in the EU over the coming weeks. The change covers eligible
[7]
OpenAI Is Adding Text Watermarks in the EU Because Regulation Works
Passing off ChatGPT's work as your own is about to get tougher -- at least if you live in the European Union. On Monday, OpenAI announced that it would start inserting an invisible watermark, imperceptible to the naked eye but identifiable by machine, into all text outputs generated by its AI
[8]
OpenAI adds invisible watermarks to ChatGPT text under EU rules
OpenAI will begin adding invisible watermarks to text generated by ChatGPT and Codex in the European Union. The rollout addresses transparency requirements under the EU AI Act. The system embeds a hidden statistical signal into model-generated writing. Readers will not see the watermark, but
[9]
OpenAI details new text watermarking system for ChatGPT, Codex, API
OpenAI today announced how it will comply with the EU AI Act, which requires providers of generative AI systems to mark AI-generated text in a way that makes it identifiable. Here are the details. A bit of context A few months ago, Anthropic sparked controversy when it announced that, to comply
[10]
OpenAI Unveils ChatGPT Watermarking
OpenAI Unveils ChatGPT Watermarking as European Union AI Rules Take Effect OpenAI will begin rolling out technology to identify AI-generated text in the European Union as new transparency requirements under the EU AI Act take effect. The company said in a blog post that API developers worldwide
[11]
OpenAI Rolls Out Text Watermarks to Meet EU AI Act Mandates | PYMNTS.com
OpenAI built the technologies in response to the European Union's AI Act, which requires generative artificial intelligence providers to make AI-generated text identifiable in a machine-readable way, according to the article. OpenAI is enabling API customers globally to opt in to text watermarking
Share
Copy Link
OpenAI announced it will add invisible watermarks to ChatGPT and Codex text outputs in the European Union using its textGrain technology. The move complies with EU AI Act transparency rules requiring machine-readable identification of AI-generated content, though the watermark can be weakened through editing.
OpenAI Deploys Invisible Watermark for EU Compliance
OpenAI announced Monday it will begin adding an invisible watermark to text generated by ChatGPT and Codex for users in the European Union over the coming weeks
1
. The text watermarking initiative addresses the EU AI Act transparency rules that took effect on August 2, requiring AI companies to mark AI-generated content in a machine-readable way1
. The watermark will apply to eligible ChatGPT and Codex users on all plans within the EU, while API developers worldwide can opt into the feature for select models starting today, though it remains off by default1
3
. OpenAI stated it is not making text watermarking a global default at launch, taking a regional approach to learn from real-world use and feedback3
.How textGrain Technology Works
The watermark operates through OpenAI's proprietary textGrain system, which adds an invisible statistical signal to the model's word choices rather than inserting an actual visible symbol
4
5
. The system subtly shapes word selections, leaving a detectable pattern that readers cannot see but specialized detectors can identify1
. Because the watermark lives within the words themselves, it travels with the text when copied and pasted1
. OpenAI published a technical report co-written with researchers from the University of Pennsylvania and Yale, demonstrating how a secret key sorts next-word predictions to create hundreds of subtle nudges that detectors can spot using only the text and the key1
. The company claims textGrain matched or exceeded other approaches like Google DeepMind's SynthID for text, which also serves as the basis for watermarking Anthropic announced in August2
5
.Detection Limitations and Editing Vulnerabilities
OpenAI acknowledges significant limitations in its watermarking technology. In tests of 400-token passages, replacing 10% of words with synonyms reduced detection rates from approximately 92% to 66%, while replacing 25% of words dropped detection to just 17%
4
. At a 1% false-positive target, OpenAI detected the watermark in about 80% of 200-token psychology responses, compared with roughly 95% when text reached 400 tokens4
. The company admits that shorter passages, mathematics content, and translated text present particular detection challenges1
4
. OpenAI cautioned that a missing watermark does not prove human authorship, as text could be too short, heavily edited, or generated by another company's AI1
. The watermark does not identify the user, reveal prompts or conversations, or indicate how much human judgment, editing, or creativity contributed to the final work1
4
.Related Stories
Controlled Access to Detection Tools
Due to these limitations, OpenAI is restricting initial detector access to approved researchers and expert organizations who can evaluate reliability and responsible uses
1
4
. The company stated it will adapt its approach as evidence, standards, and regulatory requirements evolve, expanding detector access when results can be interpreted responsibly3
. OpenAI emphasized that strong performance under ideal conditions does not guarantee reliable detection in everyday use5
. The company plans to make the technology available in open source so others can refine it5
.Performance Impact and Industry Context
OpenAI reported seeing no meaningful change in its models' performance with watermarking enabled
1
. Across benchmarks used to assess GPT-6 Astra, the company observed no meaningful performance differences with and without watermarking3
4
. The announcement follows Anthropic's decision two months earlier to watermark text generated by Claude globally, which drew backlash from users who argued they supplied the instructions, context, and decisions while Claude served merely as a tool1
. OpenAI had previously built a text watermark but held off on releasing it, partly over concerns that users would switch to rivals without watermarking, The Wall Street Journal reported in 20241
. Article 50 of the EU AI Act mandates that providers of generative AI systems make text outputs identifiable in a machine-readable way, with pre-existing entities required to comply by December 25
. Anthropic, Google, Meta, Microsoft, and OpenAI are among companies committed to following the EU's code of practice on AI-generated content1
.References
Summarized by
Navi
[1]
[4]
Related Stories
Claude AI Rolls Out Invisible Text Watermarks as EU AI Act Deadline Approaches
03 Sept 2026•Technology
OpenAI's Cautious Approach to AI Detection Tools: Balancing Innovation and Responsibility
05 Aug 2024
AI Watermarking Defeated in Hours as Anthropic Claude's Text Labels Face Instant Workarounds
19 Aug 2026•Policy and Regulation
Recent Highlights
Recent Highlights
Today's Top Stories
Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Follow topics that matter to you and stay ahead.
