ATIBA Runs Five Paper Integrity Checks Under One Rule: the LLM Judges, It Never Invents the Evidence
arXiv 2609.04123 presents a tool running reference-integrity checking against bibliographic sources with retracted and unfindable reference flagging, venue compliance derived directly from a call-for-papers page with each verdict anchored to a verbatim quote from that page, ACM SIGSOFT Empirical Standards compliance with a hallucination defense that discards any evidence quote it cannot locate verbatim in the manuscript, multi-mode AI review via GPT-5.4 through Azure OpenAI, and verified citation suggestion. All five are built on the same principle, that an LLM is trusted to judge but never to author the evidence it judges against, which is a transferable pattern for any grounded-verification agent. A moderated study with 13 non-author participants showed 69-92% agreement (mean 85%), establishing perceived usefulness only, with objective accuracy still unmeasured.
↳ Follow the thread