The report
XRPL maintainers disclosed a failure in offer accounting that affected both accumulated balances and the invariant intended to reject unauthorized issuance. The reported proof of concept could create transferred XRP in an isolated environment; its extreme hypothetical maximum was not an observed mainnet event.
The technical case records the interacting conditions and patch chronology. The same XRPL maintainer report also documents a distinct Batch-wrapper consistency bug. The overflow report explicitly credits Veria AI; the Batch finding is credited to re-verification of Sherlock findings. They must not be merged or jointly attributed to AI research.
The overflow and the separate Batch-wrapper validation defect have different discovery provenance. The maintainer disclosure credits Veria AI for the arithmetic investigation, not for every issue described in the same report. Neither proof of concept establishes that mainnet users suffered the maximum hypothetical issuance.
An invariant that repeated the same arithmetic error
The core problem was not just a number exceeding its signed representation. The ledger invariant intended to reject unauthorized issuance reused related overflowing accumulators, so the check could agree with an incorrect state. A controlled proof of concept demonstrated that both the transition and its supposed guard could fail together. The corrected checked, wider arithmetic targets that shared failure mode. The astronomical issuance ceiling discussed in the disclosure is a modeled limit, not a mainnet loss, theft or supply change.
Discovery credit also requires precision. The offer-overflow investigation is attributed to Cayden Liao and Veria AI, whereas the Batch-wrapper report concerned separate re-verification work. This News item covers the AI-assisted arithmetic finding; it does not assign the unrelated Batch flaw to an AI researcher or describe either issue as exploited against mainnet users.