AI / NEWS BRIEF
OpenAI is making text watermarking optional for selected API models and plans an EU rollout for eligible ChatGPT and Codex output, while warning that detection is imperfect.
OpenAI says global API customers can opt in to text watermarking for selected models starting October 5, with the feature off by default. Over the coming weeks, the company plans to add an invisible watermark to eligible ChatGPT and Codex text output in the European Union. Initial access to its detector will be limited to approved researchers and expert organizations.
The company calls its technique textGrain. It changes word-choice patterns so a detector can look for a statistical signal. OpenAI explicitly warns that shorter or constrained text is harder to identify and that editing can weaken detection. A positive signal does not establish authorship, ownership, accuracy or the identity of a user, while a negative result does not prove that a person wrote the text.
Text provenance is becoming an operational requirement rather than a research concept, but this launch also demonstrates why watermark results must not be treated as conclusive evidence. Regional rollout, restricted detector access and known false-positive and false-negative risks limit immediate use. The system may help platforms investigate provenance, yet decisions about misconduct or authenticity still require other evidence and human review.
OpenAI will test visual ads during image generation and add third-party conversion and brand-suitability tooling, deepening ChatGPT’s advertising business.
Claude Frontier Academy aims to train 10,000 forward-deployed engineers by the end of 2027, reflecting the growing bottleneck between capable models and reliable deployment.