OpenAI secretly watermarks AI-generated text to fight plagiarism

OpenAI is secretly watermarking text generated by its GPT-o3 and o4-mini language models to combat plagiarism, according to AI service provider Rumi. The watermarks are designed to identify AI-generated content. Rumi discovered that the watermarks consist of special, invisible Unicode characters embedded in longer text outputs. However, the watermarks can be easily removed, suggesting a temporary anti-plagiarism measure rather than a permanent solution. OpenAI has not officially announced this feature, likely to maintain its effectiveness. Rumi notes the watermarks are easily removed, potentially giving educators a false sense of security in detecting AI-generated content.