h hoge.gg
Subscribe
BTC$67,432.18+2.34%ETH$3,521.44+1.08%SOL$178.62-0.62%BNB$612.30+0.41%XRP$0.6234-0.18%ADA$0.4521+3.12%DOGE$0.1623+1.86%AVAX$38.71-1.24%LINK$17.84+0.92%HOGE$0.00004120+4.21%
BTC$67,432.18+2.34%ETH$3,521.44+1.08%SOL$178.62-0.62%BNB$612.30+0.41%XRP$0.6234-0.18%ADA$0.4521+3.12%DOGE$0.1623+1.86%AVAX$38.71-1.24%LINK$17.84+0.92%HOGE$0.00004120+4.21%
● Security & Exploits

Nine Billion, Almost No Bugs: Halborn’s 2026 Hack Casebook

Crypto lost a record $9 billion to theft in 2026, yet of the year's biggest heists only one traced to a bug in the code. Halborn's casebook shows where the money really went.

On the morning of October 29, 2026, one carefully built transaction on Ethereum did something no smart-contract exploit had managed in crypto’s short history: it moved roughly $6.3 billion out of the largest exchange on earth. There was no reentrancy bug, no integer overflow, no mispriced oracle. The code that guarded Binance’s money ran exactly as written. The attacker simply taught the wallets to answer to someone else.

That gap, between code that passes review and money that still walks out the door, is the whole story of crypto security in 2026. It was the worst year for theft the industry has ever recorded, and almost none of the headline losses came from a flaw a code auditor could have caught. The cleanest way to read the year is through the casebook of a single firm. Halborn, the Miami security shop best known for breaking into systems before criminals do, spent 2026 publishing one breach autopsy after another, most of them for incidents it did not cause and, by its own framing, could not have stopped with an audit. Taken together, those post-mortems form the closest thing the sector has to a coroner’s ledger for the year.

A signature worth $6.3 billion

The Binance breach is where the year’s lesson lands hardest. Investigators reconstructed a patient operation rather than a lucky swing. According to the published timeline, the intruders had been inside Binance’s environment since around October 20, having compromised a machine and used an internal AI tool to pivot onto two engineers who managed the wallet stack. That gave them a token with container-deployment rights. On October 29 they deployed a container, ran a script, and reached the service that signs withdrawals.

From there the mechanics split by chain. On Ethereum the attacker leaned on EIP-7702, the May 2025 feature that lets an ordinary wallet temporarily delegate its behavior to a smart contract. By stacking fifty pre-signed 7702 authorizations into a single batched transaction, they converted a Binance hot wallet into a sweeper they controlled. Because those authorizations already sat on-chain, pausing withdrawals did nothing to stop them. On the nine other blockchains, the wallets simply sent their balances out with ordinary, valid signatures, which investigators read as proof the attacker had the signing system itself approving transfers. If you have followed our explainer on account abstraction and smart accounts, the irony is sharp: a feature designed to make self-custody safer became the lever that drained an exchange.

Blockaid chief executive Ido Ben-Natan put the scale plainly, telling reporters that “this is the biggest crypto hack in history” and that his team saw “at least $3.5 billion moved out of Binance hot wallets in under an hour,” a figure he shared with crypto.news. TRM Labs traced about $6.36 billion across ten chains; Binance said it had frozen roughly $3 billion on-chain and pledged to cover the rest from reserves rather than pass losses to customers. No smart contract was broken. The exchange was simply persuaded to sign.

Binance’s response set the template for how a modern exchange survives a ten-figure breach. It suspended withdrawals, said it had frozen roughly $3 billion on-chain with help from stablecoin issuers, and promised to make users whole from its own reserves, reopening within days. The market, numb after a year of record thefts and a sliding Bitcoin price, barely flinched. Solvency, not the stolen coins, became the real question, and Binance’s answer was that its balance sheet could absorb a loss larger than most public companies are worth.

The firm that reads the bodies

Halborn was founded in Miami in 2019 by Steven Walbroehl and Rob Behnke, bootstrapped for about three years, then raised a $90 million Series A led by Summit Partners in mid-2022. Jacques Boschung took over as chief executive in 2024 and pushed the firm toward banks and tokenized assets, while Behnke moved to chairman and Walbroehl stayed on as chief technology officer. Its public catalog lists manual contract audits alongside offensive red-team engagements, blockchain layer-one assessments, and incident response, with clients that include Coinbase, Polygon, and the Solana Foundation.

What turned Halborn into the year’s unofficial record-keeper is a blog series with a deliberately clinical name: “Explained.” Each entry dissects a specific breach, usually one Halborn had no hand in, and lays out how the money left. Through 2026 the series read like a running count of the industry’s worst days, with full write-ups on Kelp DAO, Drift, Humanity Protocol, the Syscoin bridge, BonkDAO, Ostium, Coldcard, Bitget, the October Ledger supply-chain incident, and finally Binance. You can browse the catalog and watch the pattern repeat. A firm that built its name reviewing Solidity now spends most of its public output explaining breaches that never touched a line of it.

The firm has leaned into that shift commercially. Alongside audits it runs offensive engagements and an incident-response practice, and it partnered with the forensics firm Chainalysis so that a breached client can move from detection to fund-tracing without starting cold. Under Boschung the pitch has turned toward banks and tokenized-asset issuers, the institutions now wiring real balance sheets into crypto and asking who will validate the custody and signing systems that 2026 proved were the weak point. A company that once sold confidence in code increasingly sells preparation for the day the code is not the problem.

The casebook, lined up

Put the year’s biggest losses in a single column and the thesis stops being an opinion. Below are the marquee incidents Halborn and its peers reconstructed in 2026, with the one question that matters most for anyone who paid for an audit: was the smart-contract code actually at fault?

IncidentDateLoss (USD)Primary causeCode bug?
BinanceOct 29~$6.3BSigning service compromised, EIP-7702 abuseNo
BitgetSep 24~$387MThird-party backend zero-day (supply chain)No
Liquid NetworkSep 6~$320MRange-proof cache collision in ElementsYes
Kelp DAOApr 18~$292MBridge verifier config plus RPC compromiseNo
DriftApr 1~$285MGovernance control-plane plus social engineeringNo
Ledger (via reseller)Oct 9~$93MHardware supply-chain implantNo
Humanity ProtocolJun~$36MKeys on a compromised laptopNo

Seven incidents, more than $7.7 billion between them, and exactly one with a “yes” in the final column. That ratio is the headline of 2026. The attack surface has moved off the contract and onto everything around it: the keys, the signing pipeline, the governance switches, the bridges, the third-party software, and the humans.

A record year, counted three ways

Different trackers measure crime differently, so the honest approach is to cite each to its own source. CertiK, which folds phishing and scams into its totals, put 2026 losses at about $9.2 billion, the highest it has ever recorded, with roughly $6.5 billion of that landing in October alone before about $3.03 billion was frozen or returned, across 89 separate incidents that month. TRM Labs, which counts hacks and exploits more narrowly, traced about $6.36 billion to the Binance breach by itself.

The concentration is the point. One exchange breach, in one month, outweighed every prior full year of crypto theft on record. It happened while Bitcoin sat near $64,000, down roughly $28,700 from a year earlier, proof that a falling market does nothing to slow the thieves.

MetricFigureSource
2026 total losses~$9.2 billion (record)CertiK
October 2026 losses~$6.5B gross, ~$3.47B netCertiK
Binance theft (traced)~$6.36B across 10 chains, ~20 minTRM Labs
North Korea share of 2026~$7.4B (85%)TRM Labs
Stablecoin funds frozen (Binance)~$2.7B (~42%)TRM Labs
H1 2026 funds recovered~7% of incidents, ~11% of valueGlobal Ledger
Bitcoin price, Nov 18~$64,000Fortune

The control plane: how Drift lost $285 million with clean code

Rewind to April 1 and the year’s template was already set. Drift, a Solana perpetuals protocol, lost about $285 million in one of the largest exploits Solana had ever seen. The contracts were not broken. The attackers had spent months posing as a quantitative trading firm, meeting Drift staff in person at conferences and on Telegram and depositing real money to build trust.

With that access they worked the governance control-plane. In the final week of March, as Chainalysis documented, Drift’s security council was reduced to a two-of-five multisig with the timelock removed, and a set of durable-nonce transactions was pre-signed that looked routine but carried hidden authorizations. They minted a worthless token, wash-traded it in a tiny liquidity pool so it would read as collateral worth around a dollar, and borrowed against the phantom. One admin proposal and one execution, a second apart, triggered 31 withdrawals over roughly twelve minutes. Halborn’s own write-up on the drain frames it as an operational and governance failure, the kind of thing no contract audit is scoped to examine because the contracts did precisely what their owners told them to.

The detail that makes Drift unforgettable is the fake collateral. The attackers created a token called CarbonVote, propped up its price in a thin pool, and deposited it as if it were worth real money, a trick reminiscent of the 2022 Mango Markets manipulation. At roughly $285 million it ranks as the second-largest Solana exploit on record, behind only the 2022 Wormhole bridge hack, and because Drift sat near the center of Solana’s lending plumbing, a cluster of connected protocols felt the aftershock. Every contract involved had been audited. None of that mattered.

The config, not the contract: Kelp’s $292 million bridge

Seventeen days later came Kelp DAO, and the framing hardened into a slogan. OpenZeppelin, which had audited the protocol, titled its post-mortem “$292 Million Lost, Zero Bugs Found” and wrote, flatly, that “the contracts performed exactly as written.” The liquid-restaking protocol was drained of about 116,500 rsETH, and the cause sat entirely outside the code: a single trusted verifier in the cross-chain messaging setup, combined with compromised internal RPC nodes and a denial-of-service attack that forced traffic onto the poisoned ones. The bridge was told a burn had happened that never did, and it released funds against the lie.

Kelp also showed how a breach travels. Nearly 90,000 of the stolen rsETH had been posted as collateral on lending markets, so when the token’s peg wobbled, Aave and other venues scrambled to freeze rsETH markets to contain the damage. That contagion path, receipt tokens moving fluidly between protocols, is exactly the plumbing we mapped in our piece on how receipt tokens run DeFi lending. The composability that makes the system efficient also makes one bridge’s bad day everyone’s problem.

The bridge provider eventually owned the failure. LayerZero, whose messaging layer Kelp relied on, first pointed at Kelp’s configuration, then reversed course weeks later, conceding that “we didn’t police what our DVN was securing, which created a risk we simply didn’t see. We own that,” and acknowledging that nearly half of the apps on its network used the same single-verifier pattern. Arbitrum’s security council managed to freeze tens of thousands of ETH and block a second attempted drain, a rare partial save, and only possible because that chain has a council able to intervene at all.

The backend and the hardware: Bitget and the Ledger implant

September delivered two more reminders that the dangerous code often is not yours. Bitget lost roughly $387 million on September 24. Halborn’s reconstruction, in its “Explained: The Bitget Hack” entry, describes an attacker who did not steal private keys or break a contract but compromised a third-party security product in the exchange’s backend, used it to spoof unsigned transaction data, and pushed forged instructions through Bitget’s normal approval pipeline. Bitget later confirmed a zero-day in those third-party products. Cold storage was untouched; a user protection fund absorbed the loss.

It was not the first time a trusted supplier became the breach. The 2025 Bybit theft, the largest ever until Binance, ran through a compromised wallet-software vendor whose signing interface showed staff a harmless transaction while the logic underneath was quietly swapped out. Bitget was the same lesson one layer down: the thing you installed to be safer became the way in. An audit of Bitget’s own contracts would have found nothing, because Bitget’s own contracts were never the target.

Then, on October 9, the supply chain got physical. On-chain analysts estimated about $93 million drained from 311 wallets in a 47-minute coordinated sweep, all of them protected by Ledger hardware devices. The devices had been tampered with before they reached buyers, sold through a reseller called CryptoBilis in Southeast Asia. Halborn’s analysis explains that a hidden implant monitored the device screen, captured the 24-word recovery phrase during setup, and, in Halborn’s words, “included an eSIM that allowed it to transmit the stolen recovery phrase to the attacker via LTE.” It is the largest known hardware-wallet supply-chain attack on record, and the firm’s blunt advice was to buy only from the manufacturer. For anyone weighing self-custody options after a shock like that, our mobile wallet comparison lays out the trade-offs between hardware and software custody.

Keys, laptops, and the people who hold them

The smallest marquee incident of the year was also the oldest kind of mistake. Humanity Protocol lost about $36 million in June when seven private keys were lifted off a single malware-infected laptop, keys that controlled a hot wallet and two sets of bridge-admin permissions. Halborn’s write-up on the hack is almost a parable: there was nothing exotic about the attack, just a developer machine that should never have held that much power in one place.

The human layer is where 2026’s attackers spent most of their creativity, and artificial intelligence sharpened every edge. Walbroehl, Halborn’s co-founder and chief technology officer, warned that attackers are using AI to craft “highly personalized, context-aware attacks that bypass traditional security awareness training.” The Drift con, with operatives sitting across a table from staff for months, and the Binance intrusion, which reportedly used an internal AI tool to pivot between employees, are the same story told at different scales. The cheapest exploit in crypto is still a convincing person, and the tooling to be convincing has never been better.

The exception that proves the rule: Liquid’s $320 million bug

Exactly one of the year’s giant losses was a genuine code flaw, and it is worth dwelling on because of how it breaks the pattern. On September 6, Blockstream’s Liquid Network, a Bitcoin sidechain used by exchanges, lost about $320 million, roughly 4,000 BTC, from its federation. SlowMist traced the root cause to a cache-key collision deep in the range-proof verification inside Elements, Liquid’s node software, which let the attacker mint fake L-BTC and convert it to real Bitcoin. Note what this was not: not a Solidity contract, not an audit a DeFi team skipped, but a bug in consensus-level node software. Even the year’s one “code” case was not the kind of flaw a standard smart-contract review would have been hired to find.

The aftermath turned into an ethics seminar. The attackers returned about 3,400 BTC, worth roughly $269 million, once Liquid’s bridge nodes were patched, and kept around 598 BTC, calling themselves white hats. Blockstream was not having it, responding that “taking assets without authorization and withholding their return is a crime, not responsible disclosure. It is not white-hat activity. It is theft.” The federation eventually restored full one-to-one backing about two months later. The line between a bounty and a ransom is thinner than the industry likes to admit, which is part of why formal programs matter; our bug bounty map looks at where the legitimate whitehats actually operate and what they get paid.

One enemy, most of the money

Behind the biggest numbers sits a single actor. TRM Labs reported that North Korea stole about $7.4 billion in crypto in 2026, roughly 85 percent of all hack value for the year and part of a cumulative $12.8 billion since 2017. Strip out the Binance breach and the DPRK share falls to about 45 percent, which only underlines how much one operation distorted the totals. The FBI and both Chainalysis and TRM attributed Binance to TraderTraitor, the unit also tracked as Lazarus, Jade Sleet, and APT38, citing staging patterns and Tornado Cash funding that matched the 2025 Bybit theft.

What changed in 2026 is method, not just scale. Ari Redbord, TRM’s global head of policy, described the shift bluntly after the Drift hack, telling CoinDesk that the group now puts “proxies sitting across a table from protocol employees over a period of months,” a human-intelligence operation he called unprecedented and “no longer just a remote keyboard operation.” When your adversary is a nation-state running months-long cons, a two-week code review is simply aimed at the wrong target.

The in-person cons are only half of it. Through 2026, North Korean operatives also kept landing remote jobs at crypto firms under stolen or invented identities, turning the hiring pipeline itself into an attack vector, and US prosecutors brought a string of cases against the facilitators who run laptop farms on their behalf. The money matters beyond any single victim: investigators tie these thefts to the regime’s weapons programs, which is why a Lazarus hack is treated as a national-security event rather than a civil one, and why clawing the funds back is close to hopeless once they clear the first few hops.

Why a clean audit does not mean safe

None of this makes audits worthless. It makes them narrow. An audit is a point-in-time review of a defined set of contracts; it cannot vouch for a signing server, a reseller’s warehouse, a timelock an admin removes a week later, or a developer’s laptop. CertiK co-founder Ronghui Gu summed up the mismatch for Forbes: “a protocol can pass a flawless code audit and still lose millions because of a compromised admin key.” TRM’s own first-half data made the same argument in numbers, finding that infrastructure and operational failures were only about 15 percent of incidents but accounted for roughly 76 percent of the value stolen. The audit covers the most common vector. It does not cover the most expensive one.

The table below maps the gap. Each row is an attack surface; the middle column asks whether a conventional smart-contract audit even looks there.

Attack surfaceIn a standard code audit?2026 example
Smart-contract and node logicYesLiquid (range-proof)
Governance, admin keys, timelocksPartly, often out of scopeDrift
Bridge and cross-chain verifier configNoKelp DAO
Exchange backend and signing pipelineNoBinance, Bitget
Third-party software and hardwareNoLedger, Bitget
Human layer and social engineeringNoDrift, Humanity

This is why firms like Halborn increasingly sell offensive red-team exercises, continuous monitoring, and incident response alongside the traditional review. The product that matters is no longer a clean report; it is a team that assumes the breach has already happened and asks how far it would spread.

The teams that came through 2026 cleanest treated security as a budget, not a box to tick. The approach now favored by serious projects layers several defenses at once: formal verification and fuzzing on the code, strict multisig and timelocks on governance, hardware isolation and rotation for keys, continuous on-chain monitoring, and a standing incident-response plan rehearsed before it is needed. A single large protocol audit can run well into six figures, and the biggest programs now spend millions across several firms and public contests. That sounds expensive until you set it beside $6.3 billion.

The kill switch and the recovery gap

If 2026 proved that prevention keeps failing, it also clarified what actually claws money back, and the answer is uncomfortable for anyone who believes in censorship resistance. The single most reliable kill switch is a centralized stablecoin issuer freezing funds. In the Binance case, TRM credited stablecoin issuers with freezing about $2.7 billion, roughly 42 percent of the haul, which is why Tether and Circle have quietly become the industry’s emergency brake. The mechanism is a blacklist function in the token contract; it only works while the loot still sits in that stablecoin.

Tether has turned the brake into a business line, saying it has frozen billions of dollars in USDT over its history while working with hundreds of law-enforcement agencies around the world. That power cuts both ways. In one case, Thai businessmen sued Tether over roughly $42 million frozen months before any US warrant arrived, a reminder that the most effective recovery tool in crypto is a private company deciding, on its own timetable, whose money is still spendable.

Everything else is harder. Global Ledger’s first-half review found that funds were recovered in only about 7 percent of incidents, representing roughly 11 percent of stolen value. Base-layer freezes, like the validator action that trapped funds on a few-validator chain, are impossible on Bitcoin or Ethereum mainnet by design. Negotiation only works when the thief is not a sanctioned state, which rules out most of the year’s biggest cases. Tracing has never been faster, thanks to firms like Chainalysis and independent sleuths who name attackers within hours, but tracing is not recovering. The money usually moves through bridges and no-questions-asked venues before a freeze order can land.

Rules without referees

Here is the part that surprises newcomers: no US regulator mandates a smart-contract audit or accredits the people who perform one. There is no PCAOB for Solidity, no licensing board for the firms whose stamp protocols advertise. Quality is policed by reputation alone. Under chair Paul Atkins, the SEC has leaned into digital assets through its Project Crypto effort, but that work is about token classification and market structure, not code-review standards or breach testing. When a US exchange is hacked, the real muscle comes from the FBI on attribution, the Department of Justice on seizure, and the Treasury on sanctions that give a stablecoin freeze its legal backbone.

Europe has gone a step further in one narrow sense. The Digital Operational Resilience Act, binding on crypto service providers since January 2025, does reach the kind of infrastructure and third-party failures that defined 2026, and its threat-led penetration-testing regime makes adversarial testing mandatory for the largest firms by 2028. Even DORA, though, does not certify a smart-contract auditor. The regulatory calendar is crowded regardless; our guide to the lame-duck rulebook race tracks what Washington is trying to pass before January. For now, the uncomfortable truth stands: the badge on a protocol’s website is a marketing asset, not a guarantee, and nobody is checking the checkers.

The drumbeat into 2027

The giant breaches grabbed the headlines, but the steady drip never stopped. November alone brought a run of smaller hits that rhyme with the year’s theme. Brazil’s largest exchange, Mercado Bitcoin, was drained of about $40 million on November 2 across at least ten networks, hot wallets again, with the company pledging to absorb the loss. Two weeks later the perpetuals venue Lighter paused withdrawals after an attacker used a leaked trading API key to move about $10.4 million, stressing that its contracts were never compromised. Lending market Kava Lend and a handful of smaller DeFi protocols rounded out a month in which the vector was almost always a key, a credential, or a configuration, not a bug.

None of those November hits was a billion-dollar event, and that is the quiet point. Strip out the two or three mega-breaches and 2026 looks like a normal year of steady, grinding losses, the familiar diet of leaked keys, drained bridges, and manipulated oracles. What set the year apart was the tail: a handful of operations so large they bent the annual totals into a new shape and dragged the average far above the median. Defending against the median hack and defending against the tail are not the same job, and most of the industry is still built for the median.

So what does 2027 look like from here? More of the same, pointed in two directions. AI will keep lowering the cost of both attack and defense, letting a single operator run convincing long cons while letting auditors scan more code faster. Institutions arriving with tokenized assets will demand the custody, key-management, and signing-pipeline rigor that 2026 proved was the actual battleground, which is precisely the market Halborn is chasing. And the casebook will keep growing, because as long as the money is held by people, servers, and supply chains rather than by the contracts alone, the next $9 billion year will not be stopped by a cleaner audit. It will be stopped, if at all, by treating the signature itself as the thing worth guarding.

Frequently Asked Questions

How much crypto was stolen in 2026?

CertiK, which includes phishing and scams, put total 2026 losses at roughly $9.2 billion, the highest on record, with about $6.5 billion of it in October alone. TRM Labs, counting hacks more narrowly, traced about $6.36 billion to the single Binance breach on October 29, which by itself exceeded every prior full year of crypto theft.

Did Halborn audit Binance, Bitget, or Kelp before they were hacked?

No. Halborn’s role in the year’s biggest incidents was the post-mortem, published through its “Explained” blog series, not a pre-breach audit. In most of these cases there was no smart-contract bug for an audit to find; the losses came from compromised keys, backends, bridges, governance settings, and supply chains.

What was the biggest crypto hack of 2026?

The Binance breach on October 29, 2026, at roughly $6.3 billion, is the largest crypto theft on record. Attackers compromised Binance’s withdrawal-signing infrastructure and, on Ethereum, abused the EIP-7702 feature by stacking fifty pre-signed authorizations to turn a hot wallet into a sweeper. Investigators attributed it to North Korea’s TraderTraitor group.

If a project is audited, is it safe from hacks?

No. An audit is a point-in-time review of specific code and cannot vouch for keys, signing servers, admin controls, third-party software, or staff. As CertiK’s Ronghui Gu put it, a protocol can pass a flawless code audit and still lose millions through a compromised admin key. In 2026, only one of the biggest losses, Liquid, traced to an actual code bug.

Who is Halborn and what does it do?

Halborn is a Miami-based blockchain security firm founded in 2019. It offers smart-contract audits, offensive red-team and penetration testing, blockchain layer-one assessments, and incident response, and counts Coinbase, Polygon, and the Solana Foundation among its clients. It is widely known for its “Explained” breach post-mortems.

By Anneke de Vries, senior security correspondent at HOGE Wire.

Share 𝕏 Post Telegram