September was the worst month of 2026 for crypto theft. Losses from hacks reached $766.49 million, about 462% above August's $136.3 million, according to PeckShield figures reported by BeInCrypto via Yahoo Finance. Two incidents, Bitget and Liquid Network, accounted for about $707 million of that total, or roughly 92%. The other 53 incidents counted that month cost about $59 million combined.
The headline number is large, but the more useful finding is how the biggest losses happened. In the largest cases, the core code mostly did what it was designed to do. The failures sat in who was allowed to act: stolen credentials, approved signers, permission lists, and the systems wrapped around the code. For analysts, that changes which risks deserve the closest attention.
How This Analysis Was Built
This piece reviews incidents reported between April and October 2026 and sorts them by the point of failure, not by the size of the loss. We use three groups:
1. Code faults, where a contract's logic or math was wrong and anyone could call it.
2. Verification faults, where a bridge or gateway accepted a message or proof it should have rejected.
3. Access faults, where an attacker gained the right to act, through credentials, signer keys, social engineering, permissions, or the supply chain.
Loss figures come from security firms and investigators as cited by the reporting outlets linked inline. Early estimates often change as wallets are reconciled, so we note ranges where sources disagree. Root causes are labeled as confirmed only when the affected firm or its investigators said so. Market reference prices come from a Bitcoin quote snapshot taken at 02:31 UTC on October 12, 2026.
The Largest Loss Came Through Trusted Software
Bitget's hot-wallet theft on September 24 is the clearest example of an access fault. Loss estimates moved from $351.6 million to about $388 million as transactions were reconciled across 12 wallet addresses, according to the Quartz and BeInCrypto reporting collected on Yahoo Finance. Investigators at SlowMist and Mandiant said attackers compromised third-party security products, gained privileged system access, and issued fraudulent withdrawal requests. SlowMist dated the earliest malicious activity to August 31, more than three weeks before funds left the exchange.
The same reporting states that private keys were not extracted and cold wallets were unaffected. In other words, the key-management layer held. The weak point was the software and credentials that could instruct the hot and warm wallets to pay out.
Bitget also published a proof-of-reserves report on September 29 showing 131% overall coverage, with 142% for Bitcoin and 110% for Ethereum. Those figures say something about solvency after the event. They say nothing about whether withdrawal controls could be bypassed before it. A reserve ratio is a balance-sheet measure, while this incident was an operational-control failure.
A Multisig That Worked as Designed
The October 4 drain of a vault on Base shows the same pattern in decentralized finance. According to CryptoTicker's incident analysis, about 1,783 wstETH, worth roughly $6 million at the time, left the vault after an attacker-controlled contract was placed on its permission list. The vault owner's Safe wallet removed that contract at 08:52 UTC and added it again at 08:53 UTC. Once listed, the contract used its permitted access to borrow deposits and redeem them through Aave V3.
The article states that neither Aave's core contracts nor the Base network were compromised. On-chain analyses describe a three-of-seven multisig, and both permission changes carried valid signatures from existing signers. There was no timelock between the approval and the outflow. The cause, whether stolen signer keys or a manipulated approval flow, had not been established at the time of reporting. About $31.7 million reportedly remained in the vault under the same permission structure.
Two days earlier, two other Safe wallets running a helper module around Aave V3 lost 114.09 ETH, which the same report attributes to a helper contract with overly broad permissions, not to the core pools.
Bridges Fail at the Verification Step
Bridges sit between the code and access groups. Of the eight major bridge-related attacks PeckShield counted through mid-May, which took about $328.6 million in total, most involved forged messages, forged proofs, or compromised validator keys, according to The Crypto Times. The largest, Kelp DAO on April 18, lost about $292 million after attackers forged a cross-chain message. The Verus-Ethereum bridge lost about $11.58 million because it did not check that stated transfer amounts matched payouts. IoTeX's ioTube bridge lost about $4.3 million after a validator owner's private key was compromised.
The same source notes that Kelp DAO paused key contracts within roughly 46 minutes, which reportedly prevented an estimated further $200 million in losses. That detail matters. Response time is itself a control, and it varies widely between protocols.
Code Faults Still Happen but Matter Less in Dollar Terms
Classic code faults have not gone away. On October 7, a Manhattan jury found a defendant guilty of computer fraud and money laundering over the 2021 Uranium Finance exploit, where a pair contract miscalculated its reserves by a factor of one hundred, according to CryptoTicker's coverage of the verdict. The loss is reported in a range of $53.3 million to $57.2 million depending on the source. Smaller code faults continue each week. SlowMist flagged two incidents on Polygon and Ethereum that together lost about $32,600, as reported by The Crypto Times.
These cases are frequent, but in 2026 the largest dollar losses have come from the other two groups.
The Supply Chain Is Part of the Attack Surface
The newest case is still unresolved. Ledger said it was investigating reports that customers in Southeast Asia who bought devices from the reseller CryptoBilis lost funds, and asked the reseller to pause sales and shipments, according to Coinpedia. An on-chain investigator estimated $86.96 million drained across 98 addresses, a figure Ledger had not confirmed. The root cause has not been established, and the reports do not show a flaw in Ledger devices themselves. What the case does show is that the path from manufacturer to user is a control point that analysts rarely model.
A Worked Comparison
Putting the cases side by side, using figures from the sources above:
1. Bitget, about $388 million: access fault through compromised third-party software and credentials. Keys and cold wallets held.
2. Kelp DAO, about $292 million: verification fault through a forged cross-chain message.
3. Base vault, about $6 million: access fault through valid multisig signatures and a permission list with no delay.
4. Verus-Ethereum, about $11.58 million: verification fault through a missing amount check.
5. Uranium Finance, $53.3 million to $57.2 million in 2021: code fault in reserve math.
In the three largest 2026 cases on this list, the failure was an access or verification fault rather than a code fault. To put Bitget's loss in market terms, about $388 million is roughly 0.02% of Bitcoin's market value of about $1.67 trillion in the October 12 quote snapshot. The loss is small against the market, but large against one firm's protection fund, which Quartz reported fell from more than $464 million to under $200 million before being rebuilt to over $300 million.
What the Data Does and Does Not Tell Us
The data supports a narrow claim: in 2026's largest reported incidents, attackers more often gained the right to act than broke the code. Controls around signers, credentials, permission changes, and vendors carried the most weight.
The data does not show that code audits are unimportant. Survivorship matters here, since well-audited code may simply be failing less often. Incident totals are also concentrated, so one or two events can swing a month by hundreds of percent. Several root causes, including the Base vault and Ledger cases, are still unconfirmed, and loss estimates for most incidents shifted after first reports. Classifications here should be read as provisional.
Why This Matters for Analysts and Funds
Risk reviews that focus mainly on smart-contract audits may miss the more costly failure points. A more complete review asks a few direct questions:
1. Who can change permissions, and how many signers must agree?
2. Is there a timelock between approval and execution?
3. Which third-party tools can trigger withdrawals, and how are they monitored?
4. How fast can the team pause contracts, and has it done so before?
5. How are hardware and keys sourced and distributed?
Reserve reports, audit badges, and total value locked are useful, but none of them answer these questions.
Explore Presolt
Presolt brings token, chain, and market data into one research workflow, with editorial built to help analysts read risk in context. Explore Presolt to track the networks and protocols you follow.
Compliance Disclaimer
Presolt provides data and editorial for informational purposes only. Nothing here is investment advice or a recommendation to buy or sell any asset. Incident figures are drawn from third-party reports and may change as investigations continue. Crypto assets are volatile and you can lose your capital.