OpenAI begins adding watermarks to text
OpenAI begins adding watermarks to text
OpenAI is rolling out text watermarks, initially as an opt-in API option, with plans to extend them to ChatGPT and Codex in the EU.
OpenAI begins adding watermarks to text
OpenAI has started introducing watermarks for generated text. For now, this is an opt-in option for the API. In a few weeks, watermarking is expected to be rolled out for ChatGPT and Codex in the European Union.
Watermarking is not a perfect approach. A watermark can be detected with only about 80% accuracy on texts longer than 200 tokens, and the detector also produces around 1% false positives. Effectiveness depends on the type of content: in mathematical text, there are fewer ways to rewrite the same result differently, so accuracy drops noticeably. The same applies to code.
Why it matters
AnalysisThis matters because watermarks could change how AI-generated text is identified, but their usefulness is limited by text length, content type, and false-positive risks.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
OpenAI GPT (Flagship Models of the GPT Series)
The primary universal line of large language models from OpenAI (GPT-4, GPT-4o). Optimized for complex text analysis, programming, creativity, and daily intellectual tasks.
ChatGPT Updates: Plugin Development and Codex Workflows
Comprehensive engineering guide to creating, packaging, testing, and submitting plugins for ChatGPT and Codex: manifest specification, MCP servers, lifecycle hooks, custom marketplaces, and OpenAI review.
Codex Skills for Beginners: Automating AI Agent Workflows
A comprehensive hands-on guide to building and executing Codex Skills: SKILL.md anatomy, explicit vs. auto-triggering, skills vs. scripts vs. tools, Skills API, and safety boundaries.
Harvard Launches AI Safety Graduate Course with OpenAI and Anthropic Speakers
Harvard is offering a graduate AI safety course this fall, featuring researchers from OpenAI and Anthropic and free lecture recordings on YouTube.