Sakana AI proposes LLM-assisted peer review framework that catches planted errors in papers

Sakana AI has introduced a benchmark and a review framework, Multi-Layered Review, for testing whether LLMs can help reviewers catch errors in research papers.

The benchmark automatically plants contradictions into papers by adding statements that conflict with other content in the manuscript, and the system is modeled on the Three-Pass Approach to reading. According to Sakana AI, its system detected more errors than the other review systems it tested, including on papers withdrawn for real mistakes, while its paper-quality assessments stayed broadly consistent with human judgments. The work has been accepted at TMLR, and Sakana AI frames the tool as support for human reviewers rather than a replacement for them. These results are the company's own evaluations and have not been independently verified.