There’s already some work looking into this[1]. The authors add invisible prompts in papers/grants to embed watermarks in reviews and then show that they can de

There’s already some work looking into this[1]. The authors add invisible prompts in papers/grants to embed watermarks in reviews and then show that they can detect LLM generated reviews with reasonable accuracy (more than chance, but there’s no 100% detection yet). [1] Rao et al., Detecting LLM-Generated Peer Reviews https://arxiv.org/pdf/2503.15772 https://arxiv.org/pdf/2503.15772