ChatGPT Watermark Comes to the EU, and OpenAI Has a Detector Ready

Key Takeaways

• OpenAI is adding a ChatGPT watermark to chatbot text and coding output in the European Union to comply with the EU AI Act.
• The watermark is invisible and lives in the wording itself, leaving no hidden characters or odd punctuation.
• The detector can catch about 66% of unedited answers with roughly 150 words, rising to about 92% at around 300 words.
• Light editing weakens the watermark, and rival AI tools can also evade detection.
• The detector is not yet available to the public and is only available to approved researchers and expert organizations.

Market & Token Impact

• The EU AI Act has significant implications for the crypto industry, with criminal AI adoption climbing 40% year-on-year.
• The gap between what regulators demand and what detectors can prove will likely shape how every major AI lab handles compliance.
• OpenAI's GPT-6 Astra model is not affected by the watermark, as its benchmark scores remain unchanged.

Broader Context & What's Next

• The rollout follows the EU's August start of AI Act transparency enforcement.
• OpenAI is expanding its approach to content provenance to include text in response to EU regulatory requirements.
• The company recognizes the significant limitations of current text watermarking technology.
• The future of AI compliance will likely involve a balance between technological advancements and regulatory requirements.