# Bitget Exchange Loses ~$351.6M (On-Chain: ~$356.9M) in Suspected North Korean (TraderTraitor) Backend Compromise and Authorization-Flow Abuse

> On September 24, 2026 at 18:31 UTC, cryptocurrency exchange Bitget detected unauthorized transfers of approximately $351.6 million from a limited number of its hot and warm wallets; independent on-chain trackers (Lookonchain) later tallied ~$356.86 million across Ethereum, XRP Ledger, Arbitrum, Avalanche, Optimism, BSC and Base. CEO Gracy Chen disclosed that the attacker compromised a critical backend system in Bitget's wallet infrastructure, used it to spoof transaction data, and triggered the exchange's own authorization process to move funds out - private-key compromise and forged user withdrawals were ruled out. Attribution to North Korean state-sponsored actors (TraderTraitor / Lazarus cluster) is preliminary: some related IP addresses matched a VPN service used by a North Korean hacker organization, and IP behavior patterns and on-chain analysis are 'highly consistent' with DPRK-linked groups. Cold wallets and customer balances were unaffected, withdrawals remain suspended pending investigation, and Mandiant and SlowMist are investigating alongside law enforcement, Binance and Bybit.

- **Published:** 2026-09-25T00:00:00Z
- **Last reviewed:** 2026-09-28T18:47:09.214Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-2650
- **Card image:** https://intel.threadlinqs.com/og/threat/threat-intel-tradertraitor-bitget-exchange-breach-tl-2026-2650.webp?v=1686dc2c (Threadlinqs threat intel: Suspected TraderTraitor Bitget Hack: Detection & IOCs (TL-2026-2650) — Critical severity, 29 IOCs, 9 detection rules.)
- **ID:** TL-2026-2650
- **Severity:** CRITICAL
- **Category:** APT
- **Status:** ACTIVE
- **Actor:** TraderTraitor (North Korea)
- **Detections:** 9 (1 shown free below) · **IOCs:** 29 (2 shown free below)

## Description

Bitget, one of the world's top-10 cryptocurrency exchanges by volume with more than 125 million users, disclosed on September 25, 2026 that suspected North Korean threat actors had stolen approximately $351.6 million from a limited number of its hot and warm wallets. The unauthorized transfers were first flagged by Bitget's own security systems at 18:31 UTC on September 24, 2026, with funds still observed moving as late as 21:41 UTC. CEO Gracy Chen described the mechanism during a live Q&A and in X posts: the attacker 'compromised a critical backend system within our wallet infrastructure, used it to spoof transaction data, and triggered our authorization process to move funds out.' She likened the technique to slipping forged withdrawal slips through a bank's teller window - 'the vault keys never left the building; someone got into the office that prepares the slips, created paperwork that looked official, and sent it through the same approval window the bank uses every day.' Private-key compromise was ruled out, no forged user withdrawal requests were involved (the attackers 'directly infiltrated the platform's system and transferred funds'), and Bitget does not believe the incident was caused by internal personnel. Preliminary findings point to a compromise of third-party software in the backend wallet stack - Pluang's reporting quotes the CEO that attackers 'likely breached a commonly used tool' and 'manipulated a service that generated false transfer data, which then triggered a signing machine to move funds.' The specific method of system intrusion remains under active investigation and a full technical report was promised.

Independent on-chain analysts (DCF GOD, who first raised the alarm; Bubblemaps; Arkham Intelligence / Emmett Gallic; PeckShield; Hacken) documented two parallel waves of exfiltration. On the EVM side, roughly $183 million in ETH, USDT, USDC, AVAX, BNB, XAUT and other tokens moved from Bitget-labeled wallets across Ethereum, Arbitrum, Avalanche, Optimism, BSC and Base into a single brand-new address, 0x770b10b273fC44Fe9197D6bF20F145c2e98463Ee, over about an hour, within roughly 30 minutes of the first wave receiving ~34.75 million USDT, ~12.85 million USDC and 3,000 XAUt (gold-backed tokens), and about an hour later ~24,373 ETH from multiple tagged addresses; the address, created the day of the hack with its first transaction at ~19:05 UTC, is now tagged 'Bitget Exploiter 1' on Etherscan. On the XRP Ledger, two Bitget-labeled wallets plus a third linked account transferred a combined 102,976,680 XRP (approximately $157.5 million) in three payments beginning 19:01 UTC: 2,248,871 XRP from rGDreBvnHrX1get7na3J4oowN19ny4GzFn (labeled 'Bitget Global'), 91,420,943 XRP in a single transaction (8DF2ECF67268117A34B89BE9B452D3E888A22CBF03E5C196AD72D8B414A84195) at approximately 19:16 UTC, and 9,306,866 XRP via rwTTsHVUDF8Ub2nzV2oAeWxfJzUvobXLEf, all to a brand-new XRPL address beginning rwNhefsz. Native XRP cannot be frozen on the XRPL - freeze tools apply only to issued tokens - making the XRP portion the single least recoverable slice of the haul; the only remaining chokepoints are the exchanges and bridges the attacker must transit to cash out. Lookonchain's asset breakdown of the total (~$356.86 million) is: 102,926,478 XRP ($157.48M), 31,890 ETH ($85.75M), 34,751,168 USDT ($34.75M), 21,056,725 USDC ($21.06M), 19,668,852 USDT0 ($19.67M), 3,000 XAUt ($12.82M), 12,719 BNB ($9.88M), 821,012 AVAX ($8.38M), and 20,593,377 TRX ($7.07M). SlowMist's MistTrack tracking hub subsequently flagged 7 XRPL addresses holding ~102,926,478 XRP, 11 EVM addresses (~67,980 ETH, ~5,896 BNB, ~495 WETH, ~218,022 USDT, ~99,989 USDC), and 1 TRON address (~20,593,376 TRX).

The attacker exhibited classic anti-freeze and laundering tradecraft consistent with professional, state-linked financial theft. On Arbitrum, a fresh wallet spent $19.67 million in USDT0 to purchase 7,111 ETH in roughly six minutes via the UniswapX and 1inch Fusion aggregators, paying up to approximately 5% above market - deliberately prioritizing speed over price to convert freezeable stablecoins into ETH, which is 'decentralized and hard to seize,' before Tether or Circle could act. The consolidator address fanned funds out to six or more new wallets within minutes, a standard first laundering step, and funds were split and bridged cross-chain. On the XRP Ledger, the pile was split into five wallets (four holding 20 million XRP each, one holding approximately 22,976,677 XRP), and each hop was seeded with 0.00001 XRP dust payments from unrelated addresses - the signature spam pattern of address poisoning, where scammers hope someone copies the wrong wallet. On September 25 a 33,500 XRP test run transited the Bridgers swap service, flagged by analyst Yfarmx as a test run ahead of a larger cash-out; by then only ~400,105 XRP had left one of the five split wallets while more than 99% of the stolen XRP remained dormant. Analyst Specter published a thin flow-graph link from the Bridgers-swapped XRP to an Ethereum wallet in the TraderTraitor cluster, tied to July 2026's ~$24 million AFX hack, though the connection runs through a single small wallet (~$4,300) and could reflect shared laundering services rather than shared authorship.

Attribution remains preliminary but consistently points to North Korean state-sponsored actors. Gracy Chen stated that 'preliminary investigations found that some related IP addresses matched a VPN service used by a North Korean hacker organization, and the attack patterns were similar to previous actions by North Korean hackers,' and separately that 'based on IP behavior patterns and on-chain analysis, the attack method in this incident is highly consistent with known patterns of North Korean hacker organizations.' The incident occurred roughly one week after SentinelOne publicly attributed the TraderTraitor cluster (UNC4899) to an attack on an India-based IT services company, and the reporting frames the heist alongside TraderTraitor's known operations: the February 2025 Bybit theft of ~$1.5 billion (FBI-confirmed Lazarus), the April 2026 KelpDAO LayerZero-bridge theft of ~$292 million (116,500 rsETH, attributed with high confidence by LayerZero), and the July 2026 AFX theft (~$24M). The KelpDAO precedent is a direct technical parallel: the attackers used social engineering against a LayerZero Labs developer to obtain session keys starting March 6, 2026, spent six weeks infiltrating RPC cloud infrastructure, compromised two internal LayerZero-hosted RPC nodes, denial-of-service'd the external RPC node the DVN used as backup so failover landed on attacker-controlled nodes, and patched those nodes' memory to return truthful data everywhere except the DVN - which then confirmed a forged rsETH burn message ('a burn that had never occurred') and released 116,500 rsETH. Both incidents pivot on forged data feeding a legitimate authorization/signing flow: 'every transaction was individually valid' while 'system-state' was falsified. Security firm Blockaid attributed roughly $609 million of first-half-2026 losses to the TraderTraitor cluster, Chainalysis counted $1.34 billion stolen by North Korean groups in 47 heists during 2024, and Elliptic estimated North Korean hackers have stolen over $6 billion in crypto since 2017, with proceeds reportedly funding the ballistic missile program. For the Bitget incident, however, no subgroup has been formally confirmed and the investigation - with Mandiant, SlowMist, law enforcement, Binance Co-CEO (assisting with tracking, recovery and security intelligence) and Bybit's CEO (ready to help track funds) - is ongoing.

Impact and response: Bitget's cold wallets and the overwhelming majority of platform assets were unaffected; customer account balances remain accurate; deposits and trading continued during the security review. Withdrawals were temporarily suspended out of an abundance of caution and remain suspended pending investigators' confirmation (no restart timeline was committed; 'we will announce a timeline as soon as one is confirmed'). Bitget committed coverage from its User Protection Fund (5,500 BTC, worth more than $464 million), reported the incident to relevant institutions, engaged Mandiant and SlowMist as independent third-party investigators (executive Xie Jiayin also entrusted an independent third-party security team), and contacted the foundations of all affected chains; several foundations confirmed they had frozen hacker-controlled addresses, though this is only effective for issuable tokens such as USDT/USDC (precedent: Arbitrum froze ~$71M / 30,766 ETH of KelpDAO exploit proceeds in April 2026 before the attacker could move it). Attacks targeting exchanges and wallets are TraderTraitor's documented specialty per the April 2022 CISA/FBI/Treasury advisory AA22-108A: job-lure spearphishing to employees of crypto companies, trojanized Electron-based trading apps (DAFOM, TokenAIS, CryptAIS, AlticGO, Esilet, CreAI Deck), deployment of the Manuscrypt RAT, code-signing abuse, and compromised legitimate WordPress infrastructure used as C2. No CVE exists for this incident (no software vulnerability in Bitget's products was disclosed), the specific system-intrusion method and full technical report are pending, and the primary SOC exposure is monitoring the attacker-controlled addresses and laundering venues published here.

## MITRE ATT&CK

- T1195.002 Compromise Software Supply Chain
- T1204.002 Malicious File
- T1078 Valid Accounts
- T1684.001 Impersonation
- T1070 Indicator Removal
- T1090.002 External Proxy
- T1020 Automated Exfiltration
- T1565.001 Stored Data Manipulation
- T1565.002 Transmitted Data Manipulation
- T1657 Financial Theft
- T1190 Exploit Public-Facing Application
- T1068 Exploitation for Privilege Escalation
- T1036 Masquerading
- T1585 Establish Accounts
- T1585.001 Establish Accounts

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**Wallet Backend: Signed High-Value Transfer Without Matching User Intent (Forged Transfer-Data Trigger)** (`TL-2026-2650-SPL-001`)

Detects the core exploit trigger of the 2026-09-24 Bitget heist: an approval issued by the custody signing service for a high-value transfer whose originating transfer-intent record is missing or was generated by backend automation rather than a user-facing withdrawal request. The attacker compromised a critical third-party backend tool in the wallet stack, used it to spoof transaction data (CWE-345/CWE-290), and drove it through the exchange's own authorization process - every signed transaction was individually valid while the transfer records feeding the signer were falsified ('forged withdrawal slips through a bank's teller window'). This search correlates signing-service events against the transfer-intent journal in a 15-minute window and flags approvals with no intent record or with backend-originated intent.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1565.002, T1565.001, T1657
- **Data sources:** custody_signing_service_logs, wallet_transfer_intent_logs
- **Severity:** critical
- **Confidence:** medium

### Splunk SPL

```spl
index=wallet_backend sourcetype=signing_service | search action=sign_transfer | eval signed_time=_time | where amount_usd>=250000 | join type=left intent_id [ search index=wallet_backend sourcetype=transfer_intent earliest=-15m | fields intent_id intent_time intent_source user_id ] | where isnull(intent_time) OR intent_source="backend_automation" | eval signal=if(intent_source="backend_automation","backend_originated_no_user_intent","no_matching_intent_record") | table signed_time intent_id amount_usd asset dest_address initiator_service signer_key_id signal
```

### Microsoft KQL

```kql
let signingEvents = (CustodySigningEvents_CL | where TimeGenerated > ago(7d) | where action == "sign_transfer" | where amount_usd >= 250000); let intents = (CustodyIntentEvents_CL | where TimeGenerated > ago(7d) | project intent_id, intent_time=TimeGenerated, intent_source, user_id); signingEvents | join kind=leftouter intents on intent_id | where isnull(intent_time) or intent_source == "backend_automation" | project TimeGenerated, intent_id, amount_usd, asset, dest_address, initiator_service, signer_key_id, intent_source | sort by TimeGenerated desc
```

### Sigma

```yaml
title: Internal Transfer Signing Without Matched User Intent Record
id: 7b3d91f4-2c5a-4e8f-9a2d-6f1e0c8b4d73
status: experimental
description: Detects internal custody signing events for high-value transfers that carry a backend-automation origin marker or that cannot be matched to a user withdrawal-intent record, a signature of forged transfer data feeding a legitimate authorization/signing pipeline as observed in the TL-2026-2650 Bitget heist.
references:
    - https://www.coindesk.com/markets/2026/09/25/bitget-s-usd351-million-hack-happened-via-spoofed-transfers-not-private-keys-ceo-gray-chen-says
    - https://github.com/SigmaHQ/sigma
author: Threadlinqs Team
date: 2026/09/25
tags:
    - attack.impact
    - attack.t1565.001
    - attack.t1565.002
    - attack.t1657
    - attack.tool.trader_traitor
logsource:
    category: application
    product: custody
detection:
    selection_signing:
        action: sign_transfer
        amount_usd|gte: 250000
    selection_backend_origin:
        intent_source|contains: backend_automation
    condition: selection_signing and selection_backend_origin
falsepositives:
    - Scheduled rebalancing or chain-level sweeps that legitimately originate from backend automation; tune the amount threshold and allow-list known custody rebalancing jobs
    - Migration scripts executed by internal engineering during maintenance windows
level: high
```

**Known false positives:**

- Scheduled custody rebalancing and chain sweeps that legitimately originate from backend automation
- Maintenance migrations executed by internal engineering

**Tuning notes:** Threshold amount_usd at the 90th percentile of legitimate internal transfers for your environment (250k was calibrated to this heist). Maintain an allow-list of intent ids generated by approved rebalancing jobs, and tune the 15-minute join window to your intent-to-sign latency. Consider a second stage that alerts whenever a previously allow-listed backend service is observed generating transfer intent (supply-chain tampering lead).

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Custody Signing API: Service-Principal Sign-Ins From External or VPN/Proxy Addresses (locked: unlock with TL Creds)
- Trojanized Trading App Delivery or Manuscrypt/Volgmer RAT Corroborated by DPRK Cluster C2 (locked: unlock with TL Creds)
- Custody API: Automated High-Value Multi-Asset Transfer Bursts to Heist or Fresh Destinations (locked: unlock with TL Creds)
- DNS Resolution of TraderTraitor Lure Domains or DPRK Dynamic-DNS C2 Cluster (locked: unlock with TL Creds)
- New Account, Privileged Group Addition, or Log Clearing on Wallet Custody Hosts (Sigma) (locked: unlock with TL Creds)
- C2 Beaconing to DPRK Dynamic-DNS and Community-Flagged C2 Infrastructure (Sigma) (locked: unlock with TL Creds)
- Privileged Account Creation, Group Addition, or Log Clearing on Wallet Custody Hosts (locked: unlock with TL Creds)
- Backend Egress to DPRK Cluster C2 Domains and Community-Flagged Infrastructure (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (2 of 29)

Defanged, TLP:CLEAR.

- Domain: `dafom[.]dev`
- Domain: `tokenais[.]com`

Full set of 29 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Bitget Says Suspected North Korean Hackers Stole $351.6M After Backend Compromise](https://thehackernews.com/2026/09/bitget-says-suspected-north-korean.html)
- [Hackers steal $351.6 million in Bitget crypto exchange hack](https://www.bleepingcomputer.com/news/security/hackers-steal-3516-million-in-bitget-crypto-exchange-hack/)
- [Bitget's $351M Hack Happened Via Spoofed Transfers, Not Private Keys, CEO Gracy Chen Says](https://www.coindesk.com/markets/2026/09/25/bitget-s-usd351-million-hack-happened-via-spoofed-transfers-not-private-keys-ceo-gray-chen-says)
- [Bitget Hack: $157M in XRP Cannot Be Frozen](https://news.bitcoin.com/exchanges/bitget-hack-xrp-157m-cannot-be-frozen/)
- [Lookonchain: Bitget hack total ~$357M - full asset breakdown (103M XRP, 31,890 ETH, USDT, USDC, USDt0, XAUt, BNB, AVAX, TRX)](https://lookonchain.com/feeds/74105)
- [Bitget Hack Confirmed: $350M Drained (AirdropAlert, tracks EVM and XRP addresses)](https://airdropalert.com/blogs/bitget-hack/)
- [ChainCatcher: Bitget CEO live Q&A - related IPs matched a VPN service used by a North Korean hacker organization](https://www.chaincatcher.com/en/article/2292167)
- [ChainCatcher: Bitget collaborating with Mandiant and Slow Mist; UPF to cover platform-level impacts](https://www.chaincatcher.com/en/article/2292233)
- [ChainCatcher: SlowMist MistTrack tracks Bitget hacker addresses (7 XRPL, 11 EVM, 1 TRON)](https://www.chaincatcher.com/en/article/2292199)
- [Pluang/TokenPost: Bitget CEO - compromise of third-party software in backend wallet system; false transfer data triggered signing machine](https://pluang.com/en/news-feed/bitget-ceo-jelaskan-jalur-serangan-pertukaran)
- [TokenPost: Bitget hack on-chain estimate ~$356.86M incl. 102.93M XRP and 31,890 ETH](https://www.tokenpost.com/news/technology/23987)
- [Decrypt: Bitget Hack - $183M moved on-chain to a single address; DCF GOD/Bubblemaps/Arkham tracking](https://decrypt.co/379275/bitget-hack-183-million-crypto-exchange-wallets)
- [Gracy Chen (CEO) security notice confirming 18:31 UTC detection and User Protection Fund coverage](https://x.com/GracyBitget/status/2103235655879074084)
- [Gracy Chen (CEO) X post - Mandiant and Slow Mist comprehensive investigation; withdrawals suspended](https://x.com/GracyBitget/status/2103383343178154140)
- [CISA/FBI/DoS Joint Advisory AA22-108A: TraderTraitor - North Korean State-Sponsored APT Targets Blockchain Companies](https://www.cisa.gov/news-events/cybersecurity-advisories/aa22-108a)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-2650
