OpenAI, creator of prolific AI chatbot ChatGPT, has developed an AI watermarking tool with 99.9% effectiveness – but its rollout is being stalled by internal squabbles.

While the advent of the advanced AI chatbot has been lauded for its effectiveness and the main driver of the continued AI race, it also raises continued concerns about how to differentiate between human and AI-generated content.

ChatGPT’s conversational manner, ability to solve maths equations and write essays, have left academic institutions in the lurch as far as how to detect plagiarised and AI-generated work – and this is all besides the countless copyright infringement quarrels mounting against AI companies.

Since the introduction of ChatGPT, as well as countless AI image-generators, companies and researchers have been trying to develop AI detection capabilities and watermarking tools to discern AI generated content from that made by humans.

According to reports from The Wall Street Journal, OpenAI has developed this tool, which can label content generated by ChatGPT, about two years ago.

However, the transformational tool, which could put many of these AI-generated quandaries to rest, has yet to be released.

According to the report, the tool changes how tokens are selected slightly, created a watermark pattern similar to a SynthID for Text tool developed by Google.

Internal documents revealed that OpenAI’s tool is 99.9% effected on texts that contain enough new material generated by ChatGPT.

Despite developed two years ago and ready for release – ChatGPT itself was released just over two years ago in November 2022 – the watermarking tool’s launch has been stalled by internal squabbles.

Recommended reading

• OpenAI Enters Search Engine Arena with SearchGPT
• OpenAI’s Security Blunders Raise Cause for Concern
• OpenAI Criticised by Ex-Leader for Prioritising Products Over Safety

The company is reportedly mainly concerned with how the introduction of the tool with impact users.

A 2023 survey conducted by OpenAI revealed that nearly one in three loyal users of ChatGPT would be turned away by a watermarking technology since it could detect plagiarism and teaching.

Essentially, OpenAI developed a tool that could solve one the major issues associated with its hallmark technology, but is neglecting to launch it due to users wanting to continue exploiting this issue.