Veria says its AI found XRP Ledger bug that could create trillions of tokens
The September fix bypassed normal voting, and developers temporarily withheld source code. Veria says RippleX found no evidence of exploitation.
Listen to this story
The audio brief
Story brief
3 key pointsVeria’s disclosure shows how two integer-overflow defects in rippled could defeat each other’s safeguards: a multi-offer payment could undercharge its buyer, while the XRPNotCreated check could mistake the resulting supply increase for normal fees. The firm says its local exploit could mint 18 trillion XRP, but RippleX found no evidence of exploitation. Developers fixed the issue outside the ledger’s usual validator-voting process, highlighting both AI-assisted vulnerability discovery and the pressure it can place on how security patches are released.
- 01
Veria reported the flaw on September 22; the fix was merged September 23 and released September 25.
- 02
More than 80% of validators had upgraded by the fix’s release, according to Veria.
- 03
Veria says it received the bug bounty program’s maximum award of $250,000.
Veria Labs disclosed on October 10 that its AI agent found a way to create trillions of XRP through a single transaction. The flaw paired a payment-accounting error with a broken safety check in the XRP Ledger’s software. Developers shipped an emergency fix in September; according to Veria, RippleX found no evidence that anyone had exploited it.
The company’s newly published account describes the discovery, local testing and response to the vulnerability in rippled, the software that runs the ledger. Veria says it received the bug bounty program’s maximum $250,000 award. Its account places the defective code’s origins in 2015 and 2017—nearly a decade before the discovery.
Two calculations failed in the same way
XRP began with a fixed supply of 100 billion tokens. Transactions are not supposed to create more, and each transaction destroys a small fee. The vulnerable path ran through the ledger’s built-in exchange, where a payment can consume multiple trading offers rather than make one simple transfer.
The first error affected the running total used to charge the buyer. That total used an integer—a whole-number value—with a fixed storage limit. When the sum became too large, it could wrap around to a small number. Sellers still received their full payments, while the buyer was charged almost nothing.
A separate safeguard should have rejected the transaction. Called XRPNotCreated, it adds up changes in XRP balances and checks that the only net difference is the destroyed fee. But its running total had the same storage limit. The enormous increase could wrap into a negative number that looked like an ordinary fee payment.
Neither defect alone provided the complete attack. The payment error would normally be caught by the safeguard; the safeguard’s error needed the payment bug to become reachable. Chaining them allowed both calculations to misread the same transaction. Veria says that combination could create 18 trillion XRP in one transaction.
A local exploit, then human checks
Veria says its agent found both defects, worked out how to combine them and built a working exploit on a local network. The company’s author initially doubted the result, reread the proof of concept three times and brought in a founding engineer to review it. By morning, they considered the finding real.
The behavior also depended on how the software was built. Veria reports testing the official release binary and confirming that the arithmetic silently wrapped instead of crashing. Its technical explanation says a build configured to stop on overflow would instead have turned this path into a node crash, not the demonstrated token-creation exploit.
Things have changed with AI. You can’t just sneak in a critical bug patch in a regular public release process, because you’re going to get caught and reverse-engineered right away
RippleX engineer Mayukha Vadari, quoted by 24/7 Wall St.
An emergency patch without the usual vote
Veria’s disclosure gives a three-day sequence from private reporting to release:
- September 22: Veria reported the vulnerability through XRPL’s bug bounty program; it was confirmed that day.
- September 23: The fix was merged.
- September 25: The fix was released, with more than 80% of validators—the computers that verify transactions—upgraded.
Normally, changes to transaction processing go through an amendment process, with validators voting over several weeks. Veria says developers bypassed that process for the first time since its introduction more than ten years earlier. The emergency therefore changed not just the code, but the route used to put new transaction rules into operation.
24/7 Wall St. reports that developers also distributed binaries—ready-to-run software—before publishing the source code, which was shared later. That temporarily left outsiders unable to inspect the code validators were running. Vadari’s explanation identifies the competing risk: publishing the patch could help attackers locate the weakness before unpatched systems were protected.
Sources
- verialabs.comThe Biggest Hack in Crypto History That Never Happened
- 247wallst.comAI Uncovers XRP Ledger Bug: Is It a Threat or a Defense?
Reader comments
Newest comments first. Replies stay oldest first.