Asking Another Jurisdiction What It Holds Without Telling It What You Hold: Private Artifact Matching with a Ledger-Anchored Transcript

Abstract

Two authorities investigating related cases would both benefit from knowing which digital artifacts they hold in common. Neither can afford the usual way of finding out. Sending a list of hashes discloses the whole holding, including artifacts belonging to unrelated subjects, to a partner who may be under no obligation to discard it; and once sent, neither side can prove afterwards what was asked or what was answered. We apply Diffie-Hellman private set intersection over secp256k1 to this problem and add what the forensic setting actually needs, which is an auditable record: the Merkle roots of both blinded transcripts and a digest of the result are committed as a salted field vector and anchored on a permissioned ledger, so the exchange can later be shown to have happened exactly as claimed, with the artifact digests themselves never disclosed. We evaluate on 62,973 real artifacts recovered from a FAT32 medium by The Sleuth Kit and foremost. In a two-device scenario, 32,988 digests against 24,154 with a true overlap of 64, the protocol returns the intersection exactly, no false positives and no false negatives, in 8.4 s and 3.60 MiB of traffic, which is 2.06 times what a plain digest exchange would cost and discloses nothing beyond the intersection. Cost is linear: 66 bytes and 0.24 ms per item held. Anchoring the transcript adds 632 bytes of ledger state and 13 ms, and the disclosable record is 2,166 bytes and verifies in 0.96 ms. We also measure a protocol-hygiene failure that is easy to ship: reusing the blinding exponent across queries makes successive transcripts identical, letting a counterparty track a holding over time, where fresh exponents give zero overlap.