2025 年及 2026 年初的加密貨幣駭客事件,按美元計算的損失金額超越以往所有年度紀錄,總損失高達 34 億美元,涵蓋智慧合約漏洞、供應鏈被入侵、預言機操縱、私鑰盜竊,以及帶有政治動機的破壞行動,整體暴露出行業中真正致命的弱點,其實並不只是壞程式碼,而是過度集中的信任節點。
2025–2026 年加密駭客事件概況
數據難以爭辯,雖然不同機構的統計方法有所差異。
Chainalysis 估算,2025 年加密貨幣被盜總額達到 34 億美元,創歷史新高。TRM Labs 及 TechCrunch 則分別報導,金額約為 27 億美元。CertiK 則公布,2025 年上半年已發生 344 宗事故,損失 24.7 億美元,已超越 2024 年全年 19.8 億美元的淨損失。
作為對照,TRM Labs 曾計算,2024 年全年被盜金額約為 22 億美元。這意味著僅 2025 年前六個月,就已超過前一整年。
這段時期的特點,不在於事件數量,而在於「集中度」。
Immunefi 指出,2025 年第一季是加密產業史上最慘烈的一季,僅 40 宗事件便造成 16.4 億美元損失 —— 較 2024 年第一季增加 4.7 倍。單是 Bybit 和 Cetus 兩宗事故,就佔了約 17.8 億美元,約為 CertiK 統計上半年損失總額的 72%。
攻擊類型本身並沒有太大改變:智慧合約漏洞、預言機操縱、私鑰洩漏、交易所營運失誤、以及國家級網攻一應俱全。真正變化的是規模。2025 年上半年平均每宗駭客事件的金額,較前一年同期大約翻倍,損失集中在少數幾場「毀滅級」事故。
下面這些最嚴重的案例,其共同線索並不是技術複雜度,而是「信任」—— 被集中在單一私鑰、單一供應商、單一治理架構,或單一流動性場域。
延伸閱讀: Trump's 48-Hour Iran Warning: What It Did To BTC, ETH And XRP

Resolv:一次無抵押鑄造如何把穩定幣變成資產負債表危機
2026 年 3 月 22 日,有攻擊者入侵 Resolv 部署在 AWS Key Management Service 的特權私鑰,並利用該私鑰在協議的 USR 穩定幣上,授權了兩筆嚴重超額的鑄幣操作。
第一筆以約 10 萬美元的 USDC (USDC) 作為存入資產,卻鑄造了 5000 萬枚 USR;第二筆再鑄造 3000 萬枚。
合計約 8000 萬枚無抵押代幣流入 市場流通。鑄幣私鑰是一個單一外部帳戶(EOA),而非多簽地址,合約亦缺乏最高鑄幣上限、預言機檢查或金額驗證。
攻擊者將鑄出的 USR,透過 wstUSR 及其他穩定幣,轉換為約 11,400 枚 以太幣 (ETH),價值約 2400 萬至 2500 萬美元。USR 價格在 17 分鐘內於 Curve Finance 上暴跌 至約 0.025 美元,跌幅達 97.5%。
穩定幣被攻擊之所以特別致命,在於它會立刻揭穿其抵押資產究竟是真實穩健,還是脆弱不堪。
協議原本約 9500 萬美元的抵押池在技術上仍然存在,但在多出 8000 萬枚無抵押代幣後,Resolv 變成以約 9500 萬美元資產,對應約 1.73 億美元負債。包括 Aave、Morpho、Euler、Venus 及 Fluid 等 DeFi 協議,紛紛採取預防措施隔離相關風險敞口。
從漏洞發生、被迫拋售、脫鉤、資產負債缺口到市場恐慌,整個連鎖反應在一天內迅速上演。
延伸閱讀: Bitcoin's S&P 500 Correlation Just Flashed A Crash Warning
Bybit:年度定調的 15 億美元超級外洩事件
以金額來看,加密貨幣史上,沒有任何單一事件可與 2025 年 2 月 21 日發生在 Bybit 身上的事故相提並論。
鏈上調查員 ZachXBT 率先發現,該交易所的 以太坊 (ETH) 冷錢包出現超過 14.6 億美元的可疑資金外流。FBI 其後將此案歸咎於北韓的 TraderTraitor 集團(Lazarus Group 旗下的一個集團),並估算損失約為 15 億美元。
約 401,347 枚 ETH 遭竊,數量超過此前兩宗最大黑客案 —— Ronin Network 與 Poly Network —— 的合計。
這次外洩並非 Bybit 自家程式碼失誤所致。Sygnia 與 Verichains 的鑑證調查追溯到真正根源,是第三方多簽平台 Safe{Wallet} 的供應鏈遭到入侵。攻擊者早在 2 月 4 日前後入侵一名 Safe 開發者的 macOS 工作站,竊取 AWS Session Token,並於 2 月 19 日將惡意 JavaScript 注入 Safe 的網頁介面。
這段惡意程式僅在 Bybit 特定以太坊冷錢包發起交易時才會啟動。六位多簽簽名者中,有三位在未察覺被操控的情況下批准了交易。
Bybit 執行長 Ben Zhou 證實,交易所仍然具備償付能力,駭客事件前的儲備資產超過 160 億美元。72 小時內,Bybit 透過 Galaxy Digital、FalconX、Wintermute 及 Bitget 的緊急借貸補足 其 ETH 儲備。但到了 3 月 20 日,約 86% 被盜 ETH 已在近 7,000 個錢包中被兌換成 比特幣 (BTC)。
這個案例的教訓相當直接:一個場域、一次入侵、一場事故,就足以改寫全年的產業損失版圖。許多最嚴重的加密災難,恰恰發生在用戶最傾向於「以為規模代表安全」的地方。
延伸閱讀: After A $44M Hack, CoinDCX Now Faces A Fraud FIR
Sui 上的 Cetus:2.23 億美元攻擊如何凍結旗艦 DEX
2025 年 5 月,Cetus —— Sui (SUI) 網路上最大的去中心化交易所 —— 遭到攻擊,其流動性池被抽走約 2.23 億美元。根本原因是協議集中流動性數學函式庫中的一個整數溢位漏洞。
一段程式在比較數值是否達到門檻時,因一個 bit 的誤差,讓攻擊者能以存入單一代幣,卻獲得價值數百萬美元的流動性頭寸。
Sui 驗證者採取非常罕見的手段,在鏈上凍結約 1.62 億美元被盜資產;這項措施經治理投票通過,贊成率達 90.9%。約 6000 萬美元則在凍結前已經被橋接到以太坊。
Cetus 在中斷營運 17 天後恢復 上線,並以追回的資金、700 萬美元自有現金儲備,以及來自 Sui Foundation 的 3000 萬美元 USDC 貸款,重新填補流動性池。
一個旗艦級流動性場域一旦出事,整條鏈的公信力都會受創:代幣價格、鏈本身的聲譽、用戶信心,以及生態內部為了救火而不得不出手的干預,都會受到波及,衝擊遠遠超出單一協議本身。
延伸閱讀: Brazil Freezes Crypto Tax Rules
GMX:為何頂級永續合約平台仍損失逾 4200 萬美元
2025 年 7 月,GMX 因其部署在 Arbitrum 上的 V1 合約存在跨合約重入漏洞,遭到攻擊,損失超過 4200 萬美元。負責執行「減倉訂單」的函式,允許傳入一個智慧合約地址作為參數,而非限定為一般錢包地址。
在退還 ETH 的步驟中,執行流程被導向攻擊者的惡意合約,使其得以進行重入攻擊,將內部的定價數據操縱至實際市價的大約 1/57。 a threat of legal action. The attacker returned approximately $37.5 million to $40.5 million in tranches, retaining the bounty. GMX later completed a $44 million compensation plan for affected GLP holders.
The fact that funds were returned does not mean the system worked. White-hat framing, bounty offers, and partial recovery can soften the market reaction without removing the underlying security failure.
The vulnerability had been ironically introduced during a 2022 fix for a previous bug. GMX V2 was unaffected.
Also Read: Bitcoin Drops In Hours After Trump Threatens Iran Power Plants
Nobitex: When a Crypto Hack Becomes Geopolitical Warfare
In Jun. 2025, Nobitex, Iran's largest cryptocurrency exchange, was hacked for approximately $90 million across multiple blockchains including Bitcoin (BTC), Ethereum, Dogecoin (DOGE), XRP (XRP), Solana (SOL), Tron (TRX), and TON (TON).
The pro-Israel hacker group Gonjeshke Darande, also known as Predatory Sparrow, claimed responsibility.
The attack took place during active Israel-Iran military hostilities.
This was not a financially motivated theft. Stolen funds were sent to vanity burner addresses containing anti-IRGC messages with no recoverable private keys — effectively burning $90 million as a political statement.
The next day, the attackers publicly released Nobitex's entire source code, infrastructure documentation, and internal privacy R&D.
Some crypto hacks are not profit-maximizing attacks at all. They are sabotage, signaling, or cyberwarfare. That makes them different from protocol exploits in practically every dimension: motivation, method, aftermath, and the impossibility of recovery. Nobitex reported a partial resumption of operations afterward, but incoming transaction volumes dropped more than 70 percent year-over-year in early July.
Also Read: SBF Backs Trump's Iran Strikes From Prison
Abracadabra: The Exploit That Hit DeFi Borrowing Through GMX-Linked Cauldrons
On Mar. 25, 2025, an attacker drained approximately 6,260 ETH — worth about $13 million — from Abracadabra Finance's lending markets, known as cauldrons. The targeted cauldrons used GMX V2 liquidity pool tokens as collateral, and the exploit relied on a flash-loan-assisted self-liquidation technique that took advantage of state tracking errors within the gmCauldron contracts.
Stolen funds were bridged from Arbitrum to Ethereum. PeckShield was among the first security firms to flag the incident. GMX confirmed its own contracts were not affected.
Abracadabra offered a 20 percent bug bounty. This was the protocol's second major hack; a $6.49 million exploit had hit Abracadabra in January 2024.
The episode illustrates composability risk. A protocol may look secure on its own but become vulnerable through integrations and dependencies.
For DeFi users, what sits under the hood — which collateral types a protocol accepts, which external contracts it calls — matters more than the top-level brand they deposit into.
Also Read: CFTC And SEC Align On Crypto Haircuts
Hyperliquid and JELLY: Market-Structure Drama and Centralization Questions
On Mar. 26, 2025, an attacker opened a $4.1 million short position on the illiquid JELLY memecoin on Hyperliquid, alongside two offsetting long positions, then pumped the token's spot price by more than 400 percent.
When the short was liquidated, Hyperliquid's automated HLP vault inherited the underwater position, and the vault's unrealized losses reached approximately $13.5 million.
Hyperliquid's validators then force-closed all JELLY positions, settling at the attacker's original short entry of $0.0095 rather than the $0.50 that external oracles were reporting.
The maneuver was executed within two minutes and revealed that the protocol relied on only four validators per set.
The scandal here is not just the loss.
Bitget CEO Gracy Chen publicly called Hyperliquid "FTX 2.0." The protocol's total value locked collapsed from $540 million to $150 million in the following month, and the HYPE token fell 20 percent. Hyperliquid later upgraded to on-chain validator voting for asset delisting decisions.
What happens when a decentralized venue acts centrally in a crisis? That question is useful for any research audience even when the dollar loss is smaller than the biggest breaches. It exposed a credibility fault line.
Also Read: Strategy Holds 3.6% Of All Bitcoin
Meta Pool: Infinite Mint Risk and Why Low Liquidity Can Mask a Bigger Bug
In Jun. 2025, Meta Pool suffered a smart-contract exploit that allowed an attacker to mint 9,705 mpETH — worth approximately $27 million — without depositing any ETH collateral.
The vulnerability sat in the ERC-4626 mint function. The attacker bypassed the normal cooldown period through the protocol's fast unstake functionality.
But the realized loss was only around $132,000. Thin liquidity in the relevant Uniswap swap pools meant the attacker could extract only 52.5 ETH.
An MEV bot front-ran part of the attack, extracting roughly 90 ETH in liquidity that was later returned to the protocol. The 913 ETH originally staked by users remained safe with SSV Network operators.
Sometimes the bug is far worse than the realized loss. The exploit path in this case implied catastrophic theoretical damage, but poor liquidity capped extraction. That distinction matters for anyone evaluating DeFi risk, and it gives this case more depth than a simple ranking by dollar losses would suggest.
Also Read: UK Set To Block Crypto Donations
Cork Protocol: a16z-Backed, Still Exploited
On May 28, 2025, Cork Protocol was exploited for approximately $12 million. The attacker extracted 3,761 wstETH by exploiting flaws in the Cork Hook's beforeSwap logic and missing access controls.
The root cause was a lack of input validation combined with permissionless market creation without guard rails, which allowed the attacker to create a fake market using a legitimate DS token as the redemption asset.
Cork had received investments from a16z crypto and OrangeDAO in September 2024.
The takeaway is simple. Institutional investors, top-tier venture capital backing, and polished branding do not eliminate technical risk. Readers should not confuse fundraising quality with protocol safety, and audits — however thorough — are not guarantees. All contracts were immediately paused after detection, but the money was gone.
Also Read: Early Ethereum Whale Buys $19.5M In ETH
KiloEx: Oracle Manipulation as a Recurring DeFi Weakness
In Apr. 2025, KiloEx lost approximately $7 million to $7.5 million across Base, opBNB, and BNB Smart Chain after an attacker exploited an access-control vulnerability in the platform's MinimalForwarder contract. The flaw allowed anyone to call price-setting functions.
The attacker manipulated the oracle to report an absurdly low price for ETH — $100 — when opening leveraged positions, then closed at $10,000.
KiloEx offered a 10 percent white-hat bounty of $750,000. Four days later, the attacker returned all stolen funds, and KiloEx announced it would not pursue legal action.
The platform resumed after a 10-day pause and published a compensation plan for users whose trades remained open during the outage.
This is the cleanest case for explaining oracle risk. Bad pricing data can let attackers open and close positions at false values. Many exploits marketed as sophisticated are still built on old primitives — bad price feeds, predictable assumptions, poor validation. Oracle manipulation remains one of DeFi's most persistent weaknesses.
Also Read: Gold's WorstWeek Since 1983
圖案揭示了甚麼
以上 10 個個案,在運作機制、規模同動機上都各有不同,但佢哋共享同一種結構性模式。
財務損失最嚴重嘅事件——Bybit 同 Resolv——其實完全唔係由鏈上漏洞引起,而係基礎設施層面嘅失守:一宗係開發者電腦被入侵,另一宗係一把毫無防護、放喺雲端基建入面嘅鑄幣私鑰。正正因為呢啲本應「去中心化」嘅系統,實際上存在用戶以為不存在嘅中心化信任點,先導致兩次損失都變得災難性咁大。
像 Cetus 同 GMX 呢類協議層攻擊,雖然真係涉及程式碼漏洞,但實際影響範圍就係由治理層點樣應對決定——包括驗證者可唔可以凍結資金、協議有冇同攻擊者成功傾賞金和解,以及生態系其他參與者有冇即時提供緊急資金支援。
Nobitex 嚴格嚟講根本唔算係協議漏洞;嗰次係一場地緣政治破壞行動。
整體嚟睇,情況並唔樂觀。事件數量少咗,唔代表損失縮水;平均嚴重程度反而愈嚟愈高。單係北韓,就喺 2025 年嘅盜竊中佔據超過 20 億美元,比前年多咗 51%。
喺加密世界入面,最關鍵嘅安全防線,已經由鏈上邏輯轉移到鏈下基建、私鑰管理,同埋人為操作安全。
對散戶、代幣投資者,甚至協議團隊嚟講,數據都指向同一個結論:而家要問嘅,唔再係「呢個協議嘅智能合約有冇做過審計?」,而係「信任高度集中喺邊度?一旦嗰個點出事,會發生乜嘢後果?」





