Watermarks used to stop AI cheating

Artificial intelligence company Anthropic has announced that it will implement a watermarking system to show when text has been generated by AI.

Anthropic said that it will introduce the watermarking to its chatbot Claude worldwide, adding it to its newest systems as they are released as well as fitting it to existing models. The watermarking system will embed hidden patterns in text produced by AI, essentially giving that text a digital fingerprint. Anthropic says that the feature would “weave an imperceptible watermark directly into the text itself” but have not given specific details of how the watermarking process would work.

The announcement comes at a time when AI is increasingly being used by students to cheat. Studies have shown that 9 out of 10 undergraduates have used AI tools in their assessments. The Danish government recently said that students would need to defend their essays orally to defend against cheating using AI. Anthropic’s changes will enable teachers to run text through a program to determine whether or not it was AI-generated.

The general secretary of the Association of School and College Leaders, Pepe Di’Iasio, said: “Teachers will really welcome anything which helps them to detect where students are using AI and ensure a level playing field for all their learners. We’d hope that other platforms follow suit and this becomes an industry standard. Students who are determined to use AI will no doubt endeavour to find ways around any safeguards, but this is at least a step in the right direction. AI misuse is not only unfair on others, but it also means students are not actually doing the learning they need to properly understand the subject they are studying and to successfully progress in education.”