Reading Traces Under AI Pressure

AI-generated text now appears in both the manuscripts that venues receive and the reviews those manuscripts undergo. The countermeasures that venues have adopted share a vulnerability: any criterion that can be spelled out in full becomes a target that automation can learn to satisfy. The situation calls for a different kind of evidence entirely. We built and deployed a cross-disciplinary reading-trace platform that records which papers a researcher chooses to engage with and what they note about each encounter, deliberately omitting ratings and rankings. Several independent projects are moving in the same direction, and the design choices each project makes will determine whether the resulting traces remain informative or degrade into optimizable metrics.