OpenAI delays release of anti-cheating tool

OpenAI has developed a method to detect when ChatGPT is used to write essays or research papers, but the company still needs to release it. That decision results from an internal debate lasting two years, balancing the company’s commitment to transparency with the potential to deter users. One survey found nearly a third of loyal ChatGPT users would be turned off by the anti-cheating technology.

Concerns have been raised that the tool could disproportionately affect non-native English speakers. OpenAI’s spokeswoman emphasised the need for a deliberate approach due to the complexities involved. Employees supporting the tool argue that its benefits outweigh the risks, as AI-generated essays can be completed in seconds, posing a significant issue for educators.

The watermarking method would subtly alter token selection in AI-generated text, creating a detectable pattern invisible to human readers. That method is reported to be 99.9% effective, but there are concerns it could be bypassed through translation or text modifications. OpenAI is still determining how to provide access to the detector while preventing misuse.

Despite the effectiveness of watermarking, internal discussions at OpenAI have been ongoing since before ChatGPT’s launch in 2022. A 2023 survey showed global support for AI detection tools, but many ChatGPT users feared false accusations of AI use. OpenAI explores alternative approaches to address these concerns while maintaining AI transparency and credibility.