Right, but Unproven

Before a meeting, a deal or an interview, people ask an AI assistant about the person they are about to meet. On 23 September 2026 Sergei Ponomarev and Andrey Ekhmenin ran the same experiment on Andrey himself. Sergei ran the retail AI Person Scan: three questions to five AI models with live web search, answers recorded word for word with their sources. Andrey ran an ANDEKS assessment of what the public evidence could establish about the same person. Each result was frozen with a checksum before either side saw the other's, and the comparison was agreed by both sides afterwards. The models were given the person's name, role, company and profile link. One model could not identify him at all. Another linked him to older business records in a different country and a different script, which he later confirmed as his own; the answer itself contained nothing a reader could check, and the assessor working from the same public evidence could not establish the link. The paper separates what the models were given from what they found and from what could be established, shows how a true historical record was printed as a contradiction about the present, and how one model turned the absence of adverse findings into a positive reputation. It reports plainly that the scan's appendix printed a register number and an address the product had no way to tie to the person, and describes the rules that were changed the same day and the price of that change: a scan that cannot tell a person's past from a stranger's record now prints neither. Five pages, with the agreed Joint Comparison Record linked as the methodological appendix.