OpenAI introduces invisible text watermarking in the EU to comply with AI Act
OpenAI is rolling out an invisible text watermarking system called textGrain for ChatGPT and Codex in the European Union to comply with the EU AI Act, while offering an opt-in setting for global API customers.
OpenAI has announced text watermarking for ChatGPT and Codex outputs in the European Union to satisfy regulatory obligations under the EU AI Act. The company stated it will introduce watermarking for eligible users across all tiers in the EU over coming weeks. OpenAI affirmed that the system will not be a global default at launch, framing the regional approach as a way to gather feedback and learn from real-world usage.[10][4][6][2]
The invisible, machine-readable watermark is named textGrain. According to Techstrong.ai, the initiative is designed to comply with Article 50(2) of the EU AI Act, where penalties for non-compliance could surpass $2 billion. OpenAI acknowledged significant technical limitations with text watermarking, particularly that editing text reduces detection efficacy. Internal testing cited by Techstrong.ai revealed the watermark detector flags 80% of 200-word passages, but accuracy falls to 66% when 10% of words are modified and drops to 17% after 25% are altered.[5][6][3][10]
For developers, OpenAI enabled an opt-in setting allowing API customers worldwide to apply text watermarks to outputs from select models, leaving the feature off by default. Due to concerns regarding false positives and missed watermarks, OpenAI is not releasing the detection tool to the public. Instead, access will be granted on a case-by-case application basis strictly to approved researchers and expert organizations.[10][4][9][7]
Key facts
- OpenAI announced it will start watermarking eligible text from ChatGPT and Codex in the European Union over coming weeks to comply with the EU AI Act.
- The invisible, machine-readable watermarking system is named textGrain.
- Techstrong.ai reported the rollout complies with Article 50(2) of the EU AI Act under potential penalties that could exceed $2 billion.
- OpenAI testing found its detector catches 80% of 200-word passages, decreasing to 66% after 10% editing and 17% after 25% editing.
- API customers globally can choose to opt in to text watermarking for select models, leaving the feature off by default.
- Access to OpenAI's watermark detector is restricted to approved researchers and expert organizations on a case-by-case basis rather than being made publicly available.
Sources · 10 sources
- CO
Cointelegraph@CointelegraphPost on X ·
🇪🇺 NEW: OpenAI will begin watermarking eligible ChatGPT and Codex text in the EU over the coming weeks to comply with the EU AI Act. https://t.co/YKzRtoIcFi
Open source - TE
Techmeme@TechmemePost on X ·
To comply with the EU AI Act, OpenAI plans to add text watermarking for ChatGPT and Codex users in the EU and an opt-in setting for API customers globally (OpenAI) (Visit Techmeme dot com for the link and full context!)
Open source - TE
TechCrunch@TechCrunchPost on X ·
OpenAI will watermark ChatGPT and Codex text in the EU to comply with the AI Act. Editing can make the invisible marks harder to detect, it says. https://t.co/OGSYxA9KiA
Open source - WS
Wall St Engine@wallstenginePost on X ·
OpenAI will start watermarking eligible text from ChatGPT and Codex in the European Union over the coming weeks to comply with the EU 🇪🇺 AI Act, and that API customers globally can opt in to text watermarking for select models. https://t.co/I4MTZuqyRZ
Open source - TA
Techstrong.ai@TechstrongaiPost on X ·
OpenAI has begun embedding an invisible, machine-readable watermark into text generated by ChatGPT and Codex in the EU, complying with Article 50(2) of the AI Act under the threat of fines that could exceed $2 billion. That rollout stays opt-in outside Europe, and OpenAI's own testing found the detector catches only 80% of 200-word passages, falling to 66% after 10% of words are edited and 17% after 25%, which is why access remains limited to approved researchers. Read the full article on how textGrain works and why compliance is no longer optional: https://t.co/PNNta025P1 #EUAIAct #OpenAI #AIGovernance
Open source - TV
The Verge@vergePost on X ·
It's launching textGrain watermarking in the EU, but OpenAI says it won’t be the ‘global default’ yet. https://t.co/aewS2OqqNO
Open source - DE
Dexerto@DexertoPost on X ·
ChatGPT will start adding invisible watermarks to text it generates in the EU OpenAI says it can help approved organizations detect whether text was generated by its AI models https://t.co/b1GMxMEEJ8
Open source - ㆅ
ㆅ@howfxrPost on X ·
OpenAI is rolling out invisible, machine-readable text watermarking for ChatGPT and Codex, starting with users in the EU. 🧵 https://t.co/zA0U4nLmqC
Open source - TN
The New Stack@thenewstackPost on X ·
OpenAI brings text watermarking to its API -- and unlike Anthropic, it's off by default https://t.co/tmb7f1MhUK
Open source - OP
OpenAI@OpenAIPost on X ·
We're expanding our approach to content provenance to include text in response to EU regulatory requirements, while recognizing the significant limitations of current text watermarking technology. Our tools already help verify whether an image or audio file was created with our models. This work builds on those efforts to help people better understand when content may have been generated or edited with an OpenAI model. In the EU, we’ll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act. Customers using our API can turn on text watermarking for select models worldwide today.
Open source