F1The keys, not the code
VerifiedGas sweep inferredCause reported
Nothing was hacked in the usual sense: every action used a key that was allowed to do it. At 19:32–19:38 UTC on Sep 19 the attacker's main wallet 0x2dcc…1dfE received small amounts of ETH and BNB from about fifteen wallets, among them the address that minted NTX an hour later and the operator wallet that later released USDC: most likely the stolen keys' own gas money. At 20:21 an old wallet (0x1572…c362, active with the NTX contract since 2021) called the bridge converter and moved 8,721,530 FET to the main wallet. Reports attribute it to a compromised bridge authorizer key and a separate NuNet mint key.
Evidence: FET drain 0xfe12c63b…69e2; NTX mint 0xe14442f6…13ab (signed by the minter key); Crypto Times, AltcoinBuzz.
F2Mint, then sell what sells
Verified
Over the next eight hours: 408.5M NTX minted to the main wallet; 260M AGIX to the main wallet and 636M to a second wallet; 492M CGV to the second wallet; 499M WMTx from contract 0x63bb…d392. Through MetaMask Swap the main wallet sold the FET, most of the NTX and 21.7M AGIX for 699.8 ETH and 13.1 WETH; the second wallet sold WMTx, AGIX and CGV for 93.7 ETH. Prices collapsed quickly; most of the AGIX is still unsold.
Evidence: First FET sale 0x98f6e59b…6d1c; AGIX mints 0x494a123e…8fb3, 0x83e8f3b3…bced; CGV 0x789396ec…8116; WMTx 0x06e4a1b1…d26d.
F3A side door for USDC
Verified
At 13:10 UTC on Sep 20, 289,575 USDC left a separate contract (0x7ffe…566b) through one of its operator wallets and went straight to the second attacker wallet, in the same minute.
Evidence: 0xca9facda…e66f.
F4Chainflip said no. THORChain said yes, four minutes later
Direct link
Chainflip's records show five swaps to Bitcoin from fresh wallets fed by the attacker: 10,000, 75,000 and 118,015 USDC and 93.7 ETH completed for 5.549 BTC in total. The fifth, 100 ETH, was refused and refunded at 00:45 UTC on Sep 21. At 00:49 the refund address sent the same 99.9997 ETH into THORChain with the memo "=:BTC.BTC:bc1qn0zn0rc6tqn5ct565us9885fj9zthm3sgz8g0f": the Bitcoin address the refused Chainflip swap had named. That Bitcoin address received 3.31 BTC at 01:26.
Evidence: Chainflip deposits 0x005b886e…f6f8, 0xb4d87103…9e2a, 0x523d2e0d…29c3, 0x643f97ab…2903; refused 0x5b883c6d…1ef5, refund 0x203401ac…7b17; THORChain 0x866950b3…cfd0.
F5100 ETH to Monero, in thirty minutes
VerifiedRedemption inferred
The other 100 ETH took a longer road: a fresh wallet, a service that sent it back three hours later, a smart account that swapped it on Uniswap for 266,102 USDC at 04:53, Circle's CCTP to Arbitrum at 04:54, Hyperliquid's bridge at 05:12, and seven buys of FXMR, a Monero token on Hyperliquid, sent on to an account that collects FXMR from many users, most likely for redemption into Monero. By 05:23 it was gone.
Evidence: Uniswap 0xccb562e6…ea0e; CCTP burn 0xcdb9a07a…d295 (minted on Arbitrum to 0x84c8…adae); Hyperliquid account 0x84c8…adae.
F6Bitcoin, gathered and split
VerifiedOnward: not traced
All Bitcoin payouts, 8.74 BTC, were gathered at bc1qfu…a0nv by 01:40 UTC on Sep 21 and split again within hours. One 0.13 BTC piece went into a 209-input, 269-output transaction made of 0.5 BTC outputs, the shape of a CoinJoin. We have not followed the rest.
Evidence: Collector bc1qfurwaq4ca3avauvq4pcwthd78epdu73dmva0nv.
F7Oct 8–9: the second cash-out, and eight refusals
VerifiedDirect link
After two quiet weeks, about 303 ETH left the main wallet on Oct 8–9, in 1–50 ETH pieces through fresh one-time wallets. Ten pieces went to Chainflip asking for Bitcoin. Chainflip's broker refused eight of them (228 ETH) and refunded each within minutes; seven refunds went straight into THORChain, to the same Bitcoin addresses (≈ 208 ETH). Two swaps (25 ETH) went through Chainflip. Smaller pieces were swapped to USDT and USDC and routed through NEAR Intents and Circle's bridge. About 6.98 BTC from these swaps then sat unmoved until Oct 10. Part of the money reached exchange accounts; we have shared those details privately with the affected team.
Evidence: Chainflip status API (TransactionRejectedByBroker + refund tx), THORChain OUT memos, Etherscan, Blockstream balances read Oct 9 12:45 UTC. Example: 50 ETH deposit 0x5d039aa5ce764c7ac2bdd7328bd3e2430f89b30843fb38219e96dc60ba696eb8 → refused → THORChain 0x14f08b5a52444410734d2b0e7e8713844b189ea38a0199c3626af35b75e57f4b.
F8What is still there
Verified
On Oct 10 between 01:33 and 02:25 UTC the main wallet 0x2dcc…1dfe sent 144.5 ETH to four fresh wallets (20, 25, 50 and 49.5 ETH). Two swapped to USDT first (49,779 and 124,499 USDT); all four deposited into THORChain's router for Bitcoin. 0.54 ETH is left. THORChain's Bitcoin payouts (memos OUT:F91E…, OUT:926A…, OUT:EDC9… and others) were forwarded to bc1qe5pj…utp, which at 02:38 UTC was spent together with the attacker's earlier Bitcoin addresses (bc1qtztn…, bc1qxkh6…, bc1qyryz…, bc1qvh6e…) in one 12-input transaction. From there 6.00000000 BTC (02:38) and 3.00000000 BTC (04:27) went into fresh 2-of-2 multisig addresses that were emptied 15–20 minutes later, the same pattern as the round 1.00000000 BTC lots on Oct 9. Who holds the second key is not public. At 05:14 UTC the last 1.26 BTC followed into another 2-of-2 address, and at 04:34 one of the Oct 9 round 1 BTC lots moved on too: no attacker Bitcoin is left sitting at an address we can name. The second wallet 0x83f4…09c5 still holds 625.9M AGIX and 20,657 USDC, which Circle can freeze. 824M AGIX made from nothing stays unsellable at scale.
Evidence: Outgoing transactions of 0x2dcc read from the explorer's transaction list (not token logs: the wallet receives dozens of fake look-alike transfers). Bitcoin read via Blockstream / mempool.space; THORChain memos read from the OP_RETURN of each payout. Read Oct 10, 05:00–05:30 UTC.