Bridge Hacks 2026: The Race That Starts After the Theft
In 2026 you cannot reliably prevent a bridge hack; you can only win the race that begins the moment it fires. Most victims lose it, and barely one percent of stolen funds ever comes back.
Two cross-chain bridges broke within forty-eight hours of each other this week, and the only thing that separated their victims was what happened in the minutes after the exploit transaction landed. On November 18, an attacker rotated the signing key behind the Gonka bridge and left with about $1.05 million in stablecoins and freshly minted, unbacked WGNK, according to Coin Gabbar. By the morning of November 20 the money was back: the project set a twelve-hour deadline, and the thief returned the funds in a single 593.43 ETH transfer. The same morning, a different attacker drained roughly $13.2 million of a wrapped Monero token called XMR1, dumped all of it on the only exchange that listed it, and had the proceeds moving toward a mixer before most traders noticed the price had collapsed from $494 to under a dollar, per Cryptopolitan. Nobody is getting that money back.
That is the real state of bridge security in 2026. After five years and more than $2.8 billion in cumulative bridge losses, the industry has quietly stopped pretending it can prevent these attacks outright. The honest question is no longer how to make a bridge unbreakable, because the record says you cannot. The question is whether, when it breaks, you can win the race that starts the second it does. This is an analysis of that race: who runs it, how it is won and lost, and why barely one dollar in a hundred ever finds its way home.
The Year of More Hacks and Smaller Hauls
By DefiLlama‘s full-year tally, 2026 is the worst on record: roughly $8.78 billion stolen across 328 incidents, a count about 2.7 times higher than 2025 even as the median loss per hack fell from $1.07 million to about $0.51 million, as Altcoin Buzz summarized. One heist, the roughly $6.3 billion Binance theft on October 29, dominates the dollar figure. Strip it out and what remains is not a series of spectacular raids but a near-constant drizzle of smaller hits, many of them on cross-chain bridges.
Bridges carried their usual share. The year’s largest DeFi loss was the roughly $292 million drain of Kelp DAO’s rsETH bridge in April, traced to a forged cross-chain message, as CoinDesk reported. Below it sat a steady roster of smaller breaks: the Axelar-Secret Network bridge for about $4.67 million, Taiko’s ERC20 vault for about $1.7 million, and the November cluster of Ankr, Gonka, and XMR1. Blockaid’s narrower lens, 60 major onchain exploits across the first seven months of the year, put those losses at about $1.18 billion; CertiK logged a comparable escalation, counting roughly $1.26 billion stolen in the third quarter alone, per Coin Insider. However you slice the data, the direction is the same: more incidents, faster thefts, fewer giant heists, more frequent small ones.
And almost none of it comes back. The security firm Blockaid, studying those 60 major exploits, found that just $9.8 million, less than one percent of the stolen total, was ever returned to victims. All of this is unfolding during a brutal drawdown: Bitcoin traded near $61,321 on November 20, down almost 25 percent on the month, with Ether around $1,730, per Fortune. The honeypots are shrinking. The attacks are not.
Why You Cannot Simply Prevent It
A bridge is a honeypot by construction. To let value move from one chain to another, it must lock or hold real assets on one side while it authorizes claims on the other, which means a large, concentrated pot guarded by a comparatively small system. Vitalik Buterin warned years ago that there are fundamental limits to the security of bridges that hop across what he called zones of sovereignty, and 2026 kept proving him right.
The danger does not stop at the bridge itself. Because a wrapped asset circulates as collateral across lending markets, a forged mint can poison everything downstream; Kelp DAO’s rsETH, once drained, cascaded into bad debt on Aave, turning a single bridge bug into a system-wide event. That composability is exactly why Buterin argued the future would be multi-chain rather than cross-chain, with a 51 percent attack on one chain able to drain bridged assets held on another.
The bridge’s entire job is verification: deciding whether a message claiming something happened on another chain is true. When it fails, it is usually not because the cryptography broke but because the bridge believed a lie. Ben Fisch, chief executive of Espresso Systems, described the Kelp DAO case precisely to CoinDesk: the bridge worked as designed, it just “believed the wrong information.” CertiK co-founder Ronghui Gu made the complementary point to Forbes: a protocol can pass a flawless code audit and still lose millions because of a compromised key.
Put the year’s incidents side by side and the failures cluster into four doors into the same lock, the authorization that says yes to a mint or a withdrawal. The code can be perfect; if an attacker controls what the bridge treats as a valid instruction, the vault opens.
| Failure door | What goes wrong | Example |
|---|---|---|
| Stolen signing key | The attacker holds the real key and signs real-looking withdrawals | Binance signing service; XMR1 deployer key taken via a hosting-provider breach |
| Rotated or replaced key | The attacker swaps the authorized key set, then signs as the bridge | Gonka epoch BLS group key (Nov 18) |
| Replayed message | A valid, already-used proof is reused for a second payout | Ankr forged withdrawal receipts (Nov 9) |
| Forged proof the signers approve | Honest validators sign fraudulent data they cannot tell is fake | Alephium, Kelp DAO (2026); Wormhole, BNB Token Hub (2022) |
The Clock Starts the Instant the Transaction Confirms
Once prevention fails, the loss is not yet final. A hack opens a race with a handful of legs: detect the exploit, attribute the addresses, freeze whatever can be frozen, negotiate if there is leverage, and recover what is left. The attacker’s one structural advantage is raw speed. Blockaid measured a median of just 17 minutes from the exploit transaction to the first movement of stolen funds, with 22 percent of incidents seeing money move inside the first minute and 68 percent within the first hour. In four cases there was no window at all; the funds were gone before anyone could act.
That is the clock defenders are racing. Everything that follows, detection tooling, issuer freezes, Security Council multisigs, negotiation, is an attempt to act inside those 17 minutes, or to build a chokepoint downstream that still catches the money after they elapse. The uncomfortable truth of the year is that the hack is act one; the money is decided in act two.
| Stage | Who acts | 2026 reality | What it decides |
|---|---|---|---|
| Exploit confirms | Attacker | T plus 0 | Nothing yet; the money is still traceable |
| Detection | Monitoring firms | Often under 2 minutes | Whether anyone knows in time to respond |
| First laundering move | Attacker | Median 17 minutes; 22% under 1 minute | The head start |
| Freeze | Issuers, exchanges, Security Councils | Minutes, and only at a chokepoint | Most of the eventual recovery |
| Negotiation | Project and attacker | Hours to days | Returns where leverage exists |
| Recovery | Courts and coalitions | Rare | The final tally: 0.8% in 2026 |
Act One, Detection, Is Basically Solved
The surprising news of 2026 is that catching a live exploit is no longer the hard part. A layer of specialist monitors, Blockaid, Hypernative, SlowMist, the volunteer responders of SEAL 911, and on-chain sleuths like Onchain Lens, now watch mempools and contract state in real time. When the Binance signing service was hijacked on October 29, Blockaid flagged the drain in under two minutes, before any public report, and pinpointed the unauthorized signing within 11 minutes.
That breach also showed why detection has had to get this good. The attacker sent a single type-4 transaction carrying 50 authorization signatures, one for each possible nonce; when one matched, it delegated a Binance hot wallet to an attacker-controlled sweeper contract using an EIP-7702 delegation, handing over persistent, code-level control. As Blockaid described it, one unauthorized signature over the right nonce hands an attacker command of the account and every deposit that lands in it afterward. Against that, two minutes of warning is the difference between a contained incident and a clean sweep.
Hypernative says it detects 99 percent of hacks with more than two minutes of lead time; on March 5 it spotted an active exploit against Solv Protocol, which was not even a customer, and the alert let the team move funds and save an estimated $10 million. When Balancer’s pools were hit across nine chains, automated responses froze every pausable pool within minutes, protecting roughly $19.3 million. Both November bridge breaks, Gonka and Ankr, were flagged by Blockaid on X while the attacks were still in progress, according to crypto.news. Knowing is no longer the bottleneck. Doing something about it is.
Yet detection only matters if it triggers an action, and that handoff is where 2026 still leaks value. An alert that lands in a group chat at three in the morning is not a freeze. The firms that saved money, Hypernative at Solv and the automated pause at Balancer, did so because the alert was wired directly to an on-chain response rather than routed through a human who had to wake up, verify, and decide. Most long-tail bridges have no such wiring, which is why the same kind of early warning that contained the Binance sweep did nothing for Gonka’s reserves until the project fell back on a public deadline.
Act Two, the Freeze, and the Map of Chokepoints
A freeze only works where someone holds a switch, and the Binance case is the textbook illustration of both the power and the limits. Detection was instant, but in the twenty-minute gap between the first sweep and the first freeze at 5:14 p.m. Eastern, the attacker converted about $325 million of freezable stablecoins into roughly $285 million of unfreezable Ether, racing to beat the lever, as Chainalysis documented. The parts that stayed on freezable rails were caught: agencies and private firms froze roughly $2.8 billion, and the US Department of Justice later won court approval to seize more than $1.2 billion, per Crypto Briefing. The parts converted to native Bitcoin and Ether simply vanished.
Blockaid’s most damning finding is how rarely the switches get pulled at all. A stablecoin issuer blacklist was technically available in 16 of its incidents and used in zero of them; $97.2 million across 19 incidents stayed freezable but was never frozen; and $287.1 million sat in plain sight in attacker addresses holding non-issuable assets like Ether, Bitcoin, and stETH, where no freeze is possible. Of the $91.2 million that was recovered or frozen, 99 percent came through protocol-level controls, led by the Arbitrum Security Council’s $71 million freeze after the Kelp DAO exploit. The chokepoints work; they are just rarely wired up, or triggered in time.
There is an uncomfortable flip side to that power. The same blacklist that can claw back a thief’s stablecoins is a switch the issuer can throw against anyone, which is why Circle and Tether use it sparingly and why self-custody maximalists distrust freezable assets on principle. In 2026 the practical calculus inverted: the freezability that privacy advocates dislike turned out to be the single most reliable way to get stolen money back.
| Where the money lands | Who can freeze it | How well it worked in 2026 |
|---|---|---|
| Stablecoins (USDC, USDT) | The issuer (Circle, Tether) | Blacklist available in 16 incidents, used in 0 |
| Centralized exchange | Exchange compliance desk | Clawed back roughly $2.8 billion of the Binance theft |
| Pausable contract or Security Council | Protocol governance | 99% of all 2026 freezes (Arbitrum, $71M) |
| Thin market plus project leverage | The project, through pressure | Gonka returned in full; others mixed |
| Native BTC, ETH, stETH | Nobody | $287M sat unfreezable in plain sight |
Act Three, the Laundering Rail, Runs on a Stopwatch
If the freeze is the defender’s weapon, speed is the attacker’s. Bridges have overtaken mixers as the main laundering channel precisely because they let stolen value scatter across chains faster than any compliance desk can react. XMR1 is the clean example: the attacker dumped 5.2 million tokens on Hyperliquid, the only market that listed them, converted the proceeds to USDC through CoW Protocol, then swapped into Ether, the standard on-ramp to a mixer like Tornado Cash, all inside about 35 minutes, per Cryptopolitan. The wrapped token went to zero while native Monero barely moved; the loss was surgically contained to the people holding the bridge IOU, a reminder of what happens to a token when its backing disappears.
The scale behind the trend is stark. By one 2026 industry tally from Global Ledger, bridges handled about $2.01 billion of laundered funds in 2025, more than three times what passed through mixers, with multistage laundering in 99 percent of cases and a fastest first move measured at two seconds. Chainalysis maps the rhythm of large state-linked thefts into three waves: immediate layering in days zero to five, initial integration in days six to ten, and long-tail integration stretching to day 45. The Binance funds showed the automation behind it, with thousands of fresh addresses each seeded with exactly one TRX before the real money arrived, and North Korea, the firm notes, leans on bridges and Chinese-language money-movement services far more than ordinary criminals do.
The faster the money fragments, the less any downstream switch can catch, which is why the XMR1 attacker, routing toward a mixer within the hour, left nothing to freeze. That speed also blurs the line between a hack and an inside job: XMR1 began with a compromise of the team’s own hosting provider, the kind of ambiguity explored in our look at exit scams dressed up as exploits.
Act Four, the Deadline, Is the Last Lever for Self-Custody
When no technical chokepoint exists, the only tool left is leverage: making it more expensive for the attacker to keep the money than to return it. Gonka shows the move working. The project set a twelve-hour deadline, said it would work with law enforcement, exchanges, and analytics partners, and asked for 100 percent of the funds back; the attacker complied, returning the loot in that single 593.43 ETH transfer confirmed on-chain, as Coin Gabbar noted. It worked because the sum was modest, the trail was warm, and the alternative was a manhunt over a seven-figure haul.
This has hardened into a ritual: the attacker keeps a slice as a self-declared bounty and returns the rest. Blockstream’s Liquid attacker handed back 3,400 of about 4,000 Bitcoin and kept roughly 600 after the bug was patched, as Bitcoin.com reported. But a return under duress is not altruism. Ledger chief technology officer Charles Guillemet put it sharply to BeInCrypto: white hats “don’t drain a bridge and then solicit an on-chain contact.”
The economics are simple. For a seven-figure theft with a warm trail, the expected cost of laundering, chain-hopping, and dodging a coordinated freeze can exceed the discount of handing most of it back for a clean exit. For a ten-figure state-sponsored operation with an in-house laundering apparatus, it does not, which is why the Lazarus-linked Binance crew never negotiates and why the deadline tactic scales down, not up. When the thief is a nation-state, there is no one across the table to bargain with.
The Asymmetry That Decides Everything
Return to the two November bridges. Detection was identical; both were flagged in real time. The outcomes were opposite, and the reason was entirely downstream. Gonka’s loot was a wrapped token on a thin market the project could pressure, with a small enough sum to make a deadline credible. XMR1’s was dumped straight into open liquidity and routed to a mixer, with no issuer, no exchange, and no governance switch anywhere in the path. Same hack, same detection, different plumbing, and the plumbing decided who got paid.
That is the whole thesis in one number: 0.8 percent recovered across 2026. The race is lost by default. It is won only where a chokepoint happens to sit downstream and someone pulls the lever before the laundering clock runs out. Blockaid’s chief executive drew the obvious conclusion, telling reporters that “rapid containment will always beat forensics.” By the time investigators are tracing wallets, the money has usually already cleared the last switch.
Why Prevention Keeps Failing: the Repeat Offenders
The failures are not exotic, which is the most damning thing about them. In 2026 two bridges were hacked the same way twice: Verus was drained through the same entry path 66 days apart, and Allbridge’s Solana deployment was hit by a flaw its own team had already fixed elsewhere years earlier. Sergej Kunz, co-founder of 1inch, offered the blunt explanation to CoinDesk: security is often not the top priority, because teams focus on launching quickly and growing total value locked. The lessons are documented; they are simply not applied.
The reason the same bugs recur is structural, not accidental. A fix shipped to one deployment rarely gets ported to a sister chain; an audit covers the code as written, not the operational key management that actually fails; and the pressure to relaunch quickly almost always outweighs the case for slowing down to harden. Every new chain a bridge supports adds another verifier set, another key, and another message format to get wrong, so the attack surface grows even when the core contract never changes.
| Bridge loss | Approx. size | Mechanism | Outcome |
|---|---|---|---|
| Ronin (2022) | $625M | 5 of 9 validator keys | Partial; bailout plus some seizures |
| Wormhole (2022) | $320M | Forged signature | Backstopped by Jump |
| Nomad (2022) | $190M | One bad config, free-for-all | Mostly lost |
| Multichain (2023) | $130M | One operator’s keys | Lost |
| Kelp DAO (Apr 2026) | $292M | Forged LayerZero message | Coalition rebuild |
| Liquid (Sep 2026) | $320M | Forged range proof | 3,400 BTC returned |
| Binance (Oct 2026) | $6.3B | Signing service plus EIP-7702 | Roughly $2.8B frozen |
Prevention’s Real Answer Is to Shrink the Prize
The market is not waiting for bridges to get safer; it is moving away from the riskiest ones. More than $15 billion of tokenized value migrated from LayerZero to Chainlink’s CCIP during 2026, a shift Chainlink’s Johann Eid described to FinanceFeeds as a “continued flight to safety.” The deeper fix is to delete the honeypot entirely: Circle’s Cross-Chain Transfer Protocol burns USDC on one chain and mints it on another, so there is no pooled, wrapped IOU sitting around to steal in the first place.
The hardening at the top is real engineering, not just marketing. Modern designs stack independent verifiers, LayerZero’s configurable DVN model and Chainlink’s separate Risk Management Network among them, so corrupting one does not corrupt the bridge, and token standards like xERC-20 let an asset keep a single canonical issuer across chains instead of a patchwork of wrapped claims. But that repricing only hardens the top of the market. The bridges that actually broke in November, Gonka, Ankr, and XMR1, are precisely the long-tail, validator-and-key designs the capital is fleeing, run by small teams on thin budgets. The result is a two-speed system: blue-chip value consolidating onto a few heavily defended rails while everything else keeps breaking the same old ways. It mirrors the sorting happening across the exchange landscape in this drawdown, where scale and compliance increasingly decide who survives.
What Actually Shortens the Race
If 2026 has a builder’s lesson, it is that the response has to be wired up before the hack, not improvised after it. The projects that recovered real money did so because the machinery was already in place.
- Pausable contracts and kill-switches, so a drain can be halted mid-exploit rather than merely observed; Balancer’s pools froze in minutes because they were built to.
- Rate limits and circuit breakers that cap how much can leave per route, so a single forged message cannot empty the vault; OlympusDAO’s CCIP setup wrote these in explicitly.
- A Security Council with a fast multisig, the one mechanism that clawed back money at scale in 2026 (the Arbitrum freeze after Kelp DAO).
- Pre-arranged freeze relationships with stablecoin issuers and exchanges, so the call does not start cold while the clock runs.
- A real-time monitoring contract that can act, not just alert, plus a triage line such as SEAL 911, which assisted the Alephium response.
The SEC Cannot Refund You
US regulators drew clearer lines in 2026. A joint SEC and CFTC interpretation in March held that most crypto assets are not securities, as Decrypt reported, sorting the market into cleaner buckets. None of that is a recovery mechanism. Sanctions enforcement shows the limits from the other side: the Tornado Cash prosecutions established that code itself cannot be sanctioned away, but a developer can be put on trial, with Roman Storm’s case headed back to court, per The Block.
The one venue where recovery scaled was the courtroom, and only because a centralized chokepoint existed: the Binance seizure orders reached funds that had already touched exchanges and stablecoins. For a user of a self-custodial bridge, there is no deposit insurance and no chargeback. If the money reaches a mixer first, no agency is coming to reverse it, which leaves open the thorny question our feature on on-chain liability keeps circling: when an automated system hands your assets to a thief, who is actually on the hook.
On-chain insurance has not filled the gap. Dedicated cover capacity across the market actually contracted in 2026 even as losses climbed, per Coin Insider, leaving most bridge users with no policy to claim against and most protocols underinsured relative to the sums they custody. The practical result is that recourse is a function of plumbing, not paperwork: if your assets land somewhere a human can freeze them, you have a chance; if they do not, you have a receipt.
Bridging While the Market Bleeds
The drawdown offers no relief. Bitcoin near $61,000, down roughly a quarter on the month and about 30 percent below its October 2025 high, means smaller pots to steal, yet the record hack count shows attackers simply took more and smaller swings. Thin liquidity makes the aftermath worse, not better: a dumped wrapped token like XMR1 craters faster when there is less depth to absorb it. The macro backdrop, with a Fed decision looming over a fragile December, only sharpens the squeeze on the small teams running long-tail bridges.
So the practical advice for anyone moving assets across chains is unsentimental. Prefer native issuance such as CCTP, or a canonical chain bridge, over a wrapped IOU trading on a thin market. Before you bridge, ask one question: if this breaks, can anyone downstream freeze what I will be holding? A stablecoin can be frozen; native Bitcoin cannot. And assume that if a long-tail bridge fails while your funds are inside it, you are in the 99 percent that never gets made whole. The uncomfortable thesis of 2026 is that the safest assumption is that the bridge will break, and the only durable protection is standing on a rail where someone can still pull a switch after it does.
Frequently Asked Questions
How much stolen crypto actually gets recovered after a bridge hack?
Very little. Blockaid studied 60 major onchain exploits in the first seven months of 2026 and found that only about $9.8 million of $1.18 billion, under one percent, was returned to victims. Recovery happens mainly where a centralized chokepoint such as a stablecoin issuer, an exchange, or a protocol Security Council can freeze the funds before they are laundered.
Why did the Gonka hacker return the money but the XMR1 hacker did not?
Leverage and liquidity. Gonka’s stolen funds were a wrapped token the project could pressure, and the sum was small enough that a twelve-hour deadline backed by law enforcement made returning it the easier choice. The XMR1 attacker dumped everything on the only market that listed it and routed the proceeds toward a mixer within minutes, leaving no switch for anyone to pull.
What is the most common way bridges get hacked in 2026?
Authorization failures, not clever code bugs. The money is lost when the bridge’s signing authority is subverted through a stolen key, a rotated key, a replayed message, or a forged proof that honest validators approve. In each case the bridge authorizes a withdrawal or mint that it should have rejected.
Can the SEC or any regulator get my money back after a bridge exploit?
No. US regulators set policy and prosecute wrongdoers, but they do not refund hacked users and there is no deposit insurance for self-custodial bridges. The only recoveries at scale in 2026 came through courts acting on funds that had already touched centralized rails, such as the Binance seizure orders.
How fast do attackers move stolen funds?
Extremely fast. Blockaid measured a median of 17 minutes from the exploit to the first movement of funds, with 22 percent of incidents seeing money move within the first minute and 68 percent within the first hour. That speed is why detection alone is not enough and why an automatic freeze beats after-the-fact tracing.
By Daniel Voss, senior security correspondent, HOGE Wire.