blockintAI · Research blog

Rujira's FIN Exploit, Traced: 13.6 ETH In, 108 ETH Out, to a Wallet a Researcher Ties to Nine DeFi Hacks

BlockintAI · 2026-11-15T15:00:00.000Z

BlockintAI research. On-chain data was first read from Blockscout, a public Ethereum node and THORChain's Midgard API between 6:40 and 6:55 AM CST on November 13, 2026 (Ethereum block 26,406,154), and re-checked on a public Ethereum node at 7:08:59 AM CST (block 26,406,244). The five wallets still held the same balances. Times are Central (CST from November 1, CDT before). Dollar figures use an ETH price of $1,957.68 (CoinGecko, 7:08 AM CST November 13) unless a source's own figure is quoted.

On the afternoon of November 10, 2026, someone attacked Rujira Trade, the order-book exchange on THORChain's app layer. In an update the next morning, Rujira said the attacker "exploited an issue with the CCL skew calculations" in the BTC/USDC market. CCL is Rujira's custom concentrated liquidity. Rujira put the loss at around $200,000: about $175,000 from Liquidy Finance's CCL positions and $25,000 from users. It said all losses will be reimbursed, and it halted CosmWasm, the smart-contract layer, while it works on a fix.

Two days later, on November 12 at 12:43 PM CST, an independent researcher who posts as lilbagscientist published a thread on X. He says Rujira was the ninth exploit since September 4 that he can tie to one actor, with up to $8 million taken in two months. He named the Ethereum wallet where he says the Rujira money ended up: 0x911c3BF11dFc5497A53A7A04Ff7980532516f176.

We followed the Rujira money ourselves on THORChain and Ethereum, then checked every Ethereum wallet in his thread.

Key facts

Rujira exploit November 10, 2026, BTC/USDC market on Rujira Trade (FIN). Rujira estimates ~$200K lost and says it will reimburse all of it
Money in 13.60 ETH from 0x911c…f176 into THORChain, 1:48–2:08 PM CST, credited to THORChain address thor10twj…mggk
The attack window 3,738 calls from that address to one Rujira FIN contract, 3:14:28–3:38:57 PM CST
Money out 108.08 ETH paid back to 0x911c…f176, swaps recorded 3:46–3:50 PM CST. Net gain 94.49 ETH, about $193,000 at THORChain's own ETH price that afternoon
The researcher's chain of wallets Four direct ETH transfers link his Dano/Hemi, Affluent, Autheo, LOAN and Troves/Rujira wallets, September 14 to October 31
Still held About 1,384 ETH (about $2.7M) across five of the wallets on November 13

Rujira, step by step

THORChain lets you swap native coins from other chains, and its app layer can hold them as "secured assets" for use in contracts like Rujira's. THORChain's Midgard API records each swap and contract call. That makes the Rujira attack readable from start to finish.

1:48 to 2:08 PM CST, November 10. 0x911c…f176 made 11 deposits into THORChain's Ethereum router, the contract Etherscan labels "THORChain: THORChain Router v4.1.1". Together they came to 13.597 ETH. The memo on each deposit told THORChain to swap it into a secured asset and credit it to thor10twj…mggk. The two largest were 5.98741544 ETH into 0.16577933 BTC (tx 0xa4a3469be6c5e16072ecc0e11c4b47072e12e7d759eb39b5522932a65b8365fb) and 5.57305045 ETH into 0.15403892 WBTC (tx 0x15989923e8bb1d955e25f9c090d55580cc5d5794b8c708b374d956b9896af761). Smaller ones bought about 1,371 USDC, 274 USDT and a little BNB, ZEC, BCH, LTC, AVAX and RUNE.

3:14:28 to 3:38:57 PM CST. thor10twj…mggk made 3,738 calls to a single contract, thor1dwsnl…77sz. Midgard records them as Rujira FIN events. The address created three liquidity ranges of its own, made 3,587 trades and 134 "arb" calls, collected range fees 11 times, and closed its three ranges. That matches Rujira's description of a price spike in BTC/USDC "followed by a surge in trading activity across the orderbook."

3:46 to 3:50 PM CST. The address swapped everything back to ETH and sent it to 0x911c…f176. It now held far more than it had put in:

Swapped out ETH paid to 0x911c…f176 Ethereum payout tx
1.86543237 BTC 66.23336749 0xcbfb4cc82c29cbc9bc099fe60f23ebcc32bc54a7a5f2e384217fb98fe5e5f238
72,463.23 USDC 35.01806474 0x54662388aad1b88b0199f7483b07b8fc69f103031d7a1e25d75231ace54f346b
0.18270572 WBTC 6.50660226 0x91e1a84eb8f1b499d86dc11174990ccd7b5f424b55ca93ecabc8339b763d573b
Seven small balances 0.32
Total 108.08

The three large payouts landed on Ethereum between 4:06 and 4:23 PM CST. Each one carries an "OUT:" memo pointing back to the THORChain swap that paid it. 13.60 ETH in, 108.08 ETH out, a gain of 94.49 ETH. At the ETH price THORChain recorded for those swaps, about $2,045, that's roughly $193,000, close to Rujira's ~$200,000 estimate.

At 11:38 PM CST the same night, the wallet sent 3.24 ETH more into the same THORChain address. There were no contract calls this time, and between 4:49 and 5:47 AM CST on November 11, 3.23 ETH came back. That morning, Rujira said CosmWasm was halted.

The researcher's nine exploits

lilbagscientist's thread lists these incidents. The dates, figures and attributions below are his. The last column is what the Ethereum wallet he names shows on Blockscout.

Date (his) Protocol His figure Wallet he names What Ethereum shows
Sep 4 Dano Finance (Cardano) ~$2M 0xAD5cd7F2AFd4b5e9e758E8658A9Dc43925B9822E First funds late Sep 3 to Sep 4 CDT, including two withdrawals (52.93 and 64.19 ETH) from an address Etherscan labels Bybit: Hot Wallet 12, and 737,745.42 newly minted USDC swapped to 300.98 ETH
Sep 7 Hemi Genesis airdrop ~$250K Same wallet More USDC and USDT on the evening of Sep 6 CDT (Sep 7 UTC), swapped to ETH on CoW Protocol
Oct 12 Autheo (Base) ~$750K 0x3048b4Ee5A070F33b73E3401829847255079E0AD 278.52 ETH from Relay: Router V3 and 68.40 ETH from NEAR Intents on Oct 12
Oct 17 LOAN Protocol (XPR Network) ~$1.1M, mostly frozen 0x8fc774aA0741b9948f61900F3d1A48061806716F Funded with 33.01 ETH on Oct 16; received 10,102.62 HEMI on Oct 23
Oct 16 Affluent (TON) ~$1M 0xf77331B2893B275A5C58F97723469837A25fa574 and 0x9406…860C See below
Oct 31 Troves (Starknet) ~$667K 0x911c3BF11dFc5497A53A7A04Ff7980532516f176 6.32 WBTC and 117,192.83 USDC swapped to 270.24 ETH, plus 68.20 ETH from an address Etherscan labels Union Chain, all on Oct 31
Oct 31 b14g (Core) ~$1.5M none on Ethereum
Nov 5 Loopscale (Solana) ~$412K 0xB7Ec7eF24FD73f3000C651CA9D0Fa45b61DcF854 and 0xb8959Aca4e4C796A3BaDA66833eAe60b1BC0bfb2 0xb895…bfb2 received 241.87 ETH from Mayan: Swift v2 Destination at 12:19 AM CST Nov 5
Nov 10 Rujira (THORChain) ~$200K 0x911c…f176 Traced above

He says the actor used AI tools to find the bugs. That's his assessment, and the affected protocols haven't confirmed that one actor is behind all nine. blockint.ai's earlier reports already list 0xAD5c…822E under the August 31 Injective binary-options exploit, and 0x3048…E0AD and the two Loopscale wallets under those incidents.

Four transfers connect his wallets

An attribution thread is a claim. A transfer on-chain is evidence. Here is what links his wallets directly, with no exchange or bridge in between:

Time From (his label) To (his label) ETH Tx
Sep 14, 9:16:47 PM CDT 0xAD5c…822E (Dano, Hemi) 0xf773…a574 (Affluent) 49.097425 0x5c5048b4886036ffbf8665d0d1ad857c05ebc22553323cbd6651dfeba93fa531
Oct 5, 2:51:35 PM CDT 0xf773…a574 (Affluent) 0x3048…E0AD (Autheo) 414.405436 0x01791176a3ba6d2919b4be928ea32e888cf9b9ae50f9fb92f344e23385bcbe2d
Oct 27, 2:30:11 AM CDT 0x3048…E0AD (Autheo) 0x8fc7…716F (LOAN) 2.0 0x9d6c82fded1fa24cd777de7b9f344be3900775ef9859969ed8e93192eab47535
Oct 31, 3:34:35 AM CDT 0x8fc7…716F (LOAN) 0x911c…f176 (Troves, Rujira) 0.05 0x75aff23ad3cdba7ea65622227e644f838af97234792cae489841ff79bfeeeaf9

The last one is the first ETH 0x911c…f176 ever received, about a minute after its first token deposit. That matches the researcher's statement that it "was funded by LOAN attacker." A second link: one wallet, 0xc703c9bf…f1e2, received 19.68 ETH from 0x3048…E0AD in October and 63.45 ETH from 0x911c…f176 on November 8 and 9. Apart from hundreds of dust payments (0.0000001 ETH) from unlabeled look-alike addresses, those are its only inflows. It still holds 82.09 ETH.

Our on-chain review identified one timing gap. Every real deposit into 0xf773…a574 arrived between September 15 and October 5, and it passed its ETH to 0x3048…E0AD on October 5, eleven days before the October 16 date the thread gives for Affluent. We found no Affluent money arriving there on Ethereum. The two Loopscale wallets have no direct transfer to the others. 0xB7Ec…F854 received 165.53 ETH from an unlabeled high-volume address and 24.70 ETH from Relay on November 1, and 0xb895…bfb2 has only its one Mayan deposit. All three active wallets did pay the same unlabeled address, 0x794dbdbd…4a94, but it has more than 5,700 transactions and looks like a service, so we don't count it as a link.

Where the money sits

Wallet ETH held, 6:51 AM and 7:08:59 AM CST Nov 13
0x3048…E0AD 561.90
0x911c…f176 358.85
0xb895…bfb2 241.87 (never sent a transaction)
0xB7Ec…F854 139.71
0xc703c9bf…f1e2 82.09
Total 1,384.43, about $2.7M

0xAD5c…822E and 0xf773…a574 are empty. Apart from the 3.24 ETH round trip, 0x911c…f176 has sent out 15.72 ETH since Rujira: 3.42 ETH to 0x794dbdbd…4a94 and 12.3 ETH to another unlabeled high-volume address, 0x3335733c…3333. All of it is ETH, which no issuer can freeze. Exchanges and bridges can still watch for it.

The thread also names 0x9276…A3e9 as a laundering wallet that deposits to exchanges. We found no transactions from it on Ethereum.

What protocols can take from this

  • Small deposit, big exit. Rujira's attacker put in 13.6 ETH and took out 108. Watching for an account whose withdrawals jump far past its deposits in one session could flag an attack like this. Rujira says it's building a check called Gleipnir that will slow secured-asset withdrawals from the app layer when it sees unusual activity.
  • Old exploit wallets fund new ones. The wallet that cashed out Rujira got its first ETH from the wallet the researcher names as the LOAN Protocol attacker, ten days before Rujira. Screening incoming funds against public incident reports costs little.
  • Audits aren't the end. Rujira says it already used AI models in its security reviews and is now expanding them, with a fresh audit before the app layer restarts.

Sources