OpenAI has developed a highly accurate GPT-4 plagiarism detector, but hesitates to release it due to concerns about its impact, potential circumvention, and user backlash. While the watermarking system can reliably identify GPT-4 output, it cannot detect other models and can be bypassed by translation. OpenAI’s previous attempt at a text detector failed due to low accuracy and false positives, highlighting the challenges in creating a reliable tool.