The Mark Travels With The Paste

On 5 October OpenAI said that text from ChatGPT and Codex in the European Union will soon carry an invisible watermark, while developers using its API anywhere can switch one on. Anthropic already marks Claude text everywhere, and says it cannot yet limit the mark by region.

The mark is a statistical pattern in word choices, and OpenAI's own tests show how easily it fades: replacing a quarter of the words with synonyms cut detection from about 92 percent to 17 percent. This episode explains where each default applies, what a mark can and cannot prove, and why a missing mark proves nothing.

In this episode, Stephen Forte covers:

• Two defaults, one company. An invisible watermark on eligible ChatGPT and Codex text in the EU only, over the coming weeks, on all plans. In the API it stays off by default worldwide, with opt-in for select models.
• How the mark works. The textGrain method nudges word choices into a faint statistical pattern that a detector can look for later. OpenAI plans to release the method as open source.
• Who gets the detector. Access is by application, starting with approved researchers and expert organizations, case by case. It is not public at launch, and it does not identify the user or reveal prompts.
• What a mark cannot tell you. It does not measure human contribution, settle ownership or responsibility, identify the user or verify accuracy, and the absence of a mark does not prove a person wrote the text.
• The rival's opposite map. Anthropic marks Claude text worldwide because, in its words, it does not yet have a durable way to scope the mark by region.
• The numbers. About 80 percent detection at 200 tokens and about 95 percent at 400 tokens, at a 1 percent false positive target. Editing erodes it: 92 percent falls to 66 percent when a tenth of the words change, and to 17 percent when a quarter do.
• The broken number. Why "did a machine write this?" is a yes or no answer the mark cannot give, and what to rely on instead.

Sources:

• OpenAI, "Our approach to EU text provenance rules," 5 October 2026: https://openai.com/index/eu-text-provenance
• OpenAI, "textGrain: Entropy-Calibrated Watermarking for Language Model Text," technical report, 5 October 2026: https://cdn.openai.com/pdf/e9508624-d767-41b6-a26d-e34ca798ada6/textgrain-entropy-calibrated-watermarking-for-language-model-text.pdf
• Anthropic, "How Claude's text watermark works," 14 August 2026: https://www.anthropic.com/news/claude-text-watermark
• European Commission AI Act Service Desk, Article 50, transparency obligations: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50
• European Commission, Code of Practice on Transparency of AI-generated Content: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content

The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.