ChatGPT plagiarists should be cautious now, as OpenAI has crafted a tool that has the ability to detect the writing output of GPT-4 with an astonishing accuracy of reportedly 99.99%. Nevertheless, the company has been wavering for more than a year on whether to actually make it available to the public.

According to TechCrunch, the company is purportedly adopting a “deliberate approach” because of “the intricate matters involved and its potential influence on the broader ecosystem beyond OpenAI”. “The text watermarking method we are developing is technologically promising, but there are significant risks that we are considering while researching alternatives. These include the susceptibility to being circumvented by malicious actors and the potential to disproportionately affect groups such as non-English speakers,” an OpenAI spokesperson stated.

The text-watermarking system operates by integrating a specific pattern into the written output of the model, which is detectable by the OpenAI tool but remains invisible to the end user. While this tool can reliably identify the writing generated by its own GPT-4 engine, it is unable to detect the outputs of other models like Gemini or Claude. What’s more, the watermark itself can be eliminated by running the text output through Google Translate, converting it to another language and then back again.

This is not OpenAI’s first endeavor in creating a text-detection tool. Last year, it silently abandoned a similar text detector that was under development due to the tool’s meager detection rate and tendency for false positives. Released in January 2023, that detector required a user to manually input sample text of at least 1,000 characters in length before it could make a determination. It managed to accurately classify AI-generated content with only 26% accuracy and wrongly labeled human-generated content as AI-derived 9% of the time. It even led one Texas A&M professor to wrongly fail an entire class for allegedly using ChatGPT on their final assignments.

OpenAI is also reportedly reluctant to release the tool out of fear of a user backlash. As per the Wall Street Journal, 69% of ChatGPT users believe that such a tool would be unreliable and likely lead to false accusations of cheating. Another 30% reported that they would readily abandon the chatbot in favor of a different model if OpenAI actually rolls out the feature. The company also fears that developers would be able to reverse engineer the watermark and create tools to nullify it.

Even as OpenAI deliberates on the advantages of releasing its watermarking system, other AI startups are hastening to release text detectors of their own, including GPTZero, ZeroGPT, Scribbr, and Writer AI Content Detector. However, given their overall lack of accuracy, the human eye remains our most effective method of identifying AI-generated content, which is not very reassuring.