After Binance's $6.3 Billion Hack, Are These Four Perp DEXs Really Safe?
- Key Takeaways: The $6.3 billion Binance hack on October 29 revealed that attackers don't need private keys—they only need to compromise internal systems to make the signing system approve withdrawals. Comparing four perpetual contract DEXs, only Lighter can withstand this type of server compromise; the other three ultimately depend on team keys and custodians for fund security.
- Key Elements:
- The Binance attacker gained access to internal services by compromising a DevOps engineer's laptop, after which the system itself issued withdrawals; on Ethereum, the attacker likely exploited EIP-7702 delegation to bypass the signing system's "known bad actions" list, and $3 billion has been frozen.
- Lighter's approximately $1 billion in assets are validated and released in batches via ZK proofs, supporting 14-day forced withdrawals and a "desert mode" full user exit; however, prices are provided by the operator, and a 4/7 security committee can skip the 21-day notice for instant upgrades—the proof verifier was changed 5 times in October alone.
- Of Hyperliquid's $6.5 billion in USDC, 86% (about $5.6 billion) is held in single-key wallets under Coinbase's name, with security dependent on Coinbase's private key custody; of the 27 on-chain validators, Hyper Foundation runs 5, accounting for 47% of staked tokens, and the node software is closed-source.
- Aster's approximately $500 million vault withdrawals must be initiated by the operator and co-signed with two Aster keys—user signatures only prove the request was initiated; the EVM vault has an hourly withdrawal cap, but the same 4/7 multisig can instantly raise the cap.
- Of Variational's approximately $290 million in USDC, any one of 17 operator keys can transfer any user pool funds without amount limits; the admin is a single wallet with no timelock, and there is no forced exit function in the contracts—representing the largest risk exposure.
- Conclusion: The DEX label does not constitute a security guarantee. Assessment should focus on "where the money is, who can move it, and whether users can exit without the team"; only Lighter supports a fund exit path that does not depend on the team.
Original Author: 0xJaehaerys
Original Compilation: AididiaoJP, Foresight News
On October 29, $6.3 billion was transferred out of Binance's hot wallets. According to Binance, a DevOps engineer's laptop was compromised, giving attackers access to internal services, and withdrawals were initiated from those internal services; there is no indication that private keys were leaked.
In other words, the attackers most likely never needed private keys at all—they simply got Binance's own systems to sign the withdrawals. Binance later discovered additional breach points, and a full post-mortem has yet to be released. $3 billion has been frozen, and Binance has committed to covering the losses.
On Ethereum, a TRM report states that the attackers got Binance's signing system to approve an EIP-7702 delegation, handing control of a hot wallet to an attacker contract. Researchers believe the signing system screens against a list of "known bad actions" rather than only allowing through a whitelist of "known good actions," which is how this relatively new authorization method slipped through. Binance has not confirmed this detail, and it does not explain the losses on Bitcoin, Solana, and Tron.
A month earlier, Bitget lost over $350 million in a similar fashion: attackers breached the backend system that feeds transaction data to signers, and withdrawals went out the door with valid signatures.
If an exchange's systems can move your money, then breaching the systems is enough. Perpetual DEXs have always claimed they can structurally avoid this kind of risk. I picked four of them and tried to break down the same three questions: Where exactly is the money? Who can move it? If the team disappears or gets hacked, can you still get out?
The answers come from on-chain reads, verified contract code, team documentation and posts, and @l2beat; incident history references public reporting. This is not an audit, nor is it a hunt for vulnerabilities. I looked at only one scenario: if the team's systems and keys were compromised the way Binance's were, what could attackers do with them, and could you still get your money back?
I am not saying these protocols have vulnerabilities or will be hacked. Off-chain protections such as MPC wallets, signing policies, and monitoring are invisible from the outside and may make these scenarios much harder. So what follows describes what the contracts allow, not predictions.
Lighter: Even if the servers are hacked, your balance can't be touched
Lighter holds about $1 billion in its Ethereum contracts: $556 million USDC, $430 million LIT (151 million tokens), plus some ETH and miscellaneous assets.
In normal operation, funds can only leave through batches verified by ZK proofs under the contract. Standard withdrawals can only go to the account owner's wallet, a binding fixed at the time of first deposit. Faster withdrawals and transfers to other accounts also require your wallet's signature. So no matter who controls Lighter's sequencer, prover, or validator keys, they cannot unilaterally withdraw balances.
You also don't have to wait for them. You can initiate a forced withdrawal directly on Ethereum; if Lighter doesn't process it within 14 days, anyone can switch the contract into "desert mode." Upgrades stop, and everyone exits using ZK balance proofs based on data already on Ethereum.
But it is not entirely trustless.
Prices are supplied by the Lighter operator as index price and mark price, and the proofs do not verify oracle signatures. Incorrect prices can produce unfair liquidations, and L2BEAT flags the same risk.
There are also two guardian keys that can pause batch processing and withdrawals. One can pause for up to 24 hours at a time, the other for up to 72 hours, and each pause requires a cooldown before it can be used again. However, neither key can block "desert mode": as long as no one processes the 14-day forced withdrawal, the contract can still enter a full exit state.
Upgrades follow a different process. Normally, a 3/5 multisig must give 21 days' public notice on Ethereum, giving users time to see what changed. But if a separate 4/7 security committee also signs, that 21-day notice can be skipped entirely and the upgrade takes effect immediately. To date, every Lighter upgrade has taken this fast lane. In October alone, the proof verifier was swapped five times. The proof verifier is the contract component that validates ZK withdrawal proofs; its frequent replacement means "how fast the code can be changed" is not a paper risk.
Historically, there was about a 4.5-hour outage during the October 10, 2025 crash. Withdrawal delays appeared after LIT launched in late December 2025. On October 17, 2026, there was a 2-hour guardian pause, which L2BEAT treated as a test. On October 20 and October 29, there were roughly 4.5-hour state update interruptions, both ending after fast-track verifier upgrades. No public reports indicate custodied funds were harmed.
My assessment: of the four, only Lighter withstands a Binance-style server breach—attackers who get into the servers still cannot withdraw balances using the servers alone. The cost is that the operator controls pricing, and multisig code changes can happen very quickly. Those are the two things you really have to trust.
Hyperliquid: Most USDC sits in a single-key Coinbase wallet and its own chain
Hyperliquid is backed by $6.5 billion in USDC, spread across three locations.
- About $5.6 billion, or 86%, in a USDC reserve wallet on HyperEVM under Coinbase's name;
- About $622 million, or 10%, in Circle's CoreDepositWallet contract;
- About $290 million, or 4.4%, still on the legacy Arbitrum bridge.
That Coinbase wallet is an ordinary address with a single private key. In other words, whether 86% of the USDC is safe depends entirely on how Coinbase safeguards that key, which cannot be seen on-chain. The Hyperliquid chain can also bypass Coinbase's private key using system transactions that require no signature, moving USDC back and forth between the Coinbase wallet and the Circle contract to keep the two balances roughly at 9:1—that is, USDC in the Coinbase wallet at about 9x that in the Circle contract.
Coinbase stakes 506,000 HYPE, about $46 million, for this role. That stake is only slashed if the wallet fails to pay the protocol's revenue share, and it is less than 1% of the balance—hardly an endorsement of the wallet.
The Circle contract is upgradeable, with a single address as admin; the Hyperliquid chain can also transfer USDC out directly. On the Arbitrum bridge, 3 of 4 validators signing is enough to release funds, with a 200-second window to intercept before release.
The backstop is Circle. It can freeze native USDC, and the $823 million USDC stolen from Binance was frozen before it could move. Freezing stops thieves, but the money stays stuck until someone resolves it.
There are 27 active validators on the chain. The 5 run by the Hyper Foundation hold 47% of the stake; the top three together exceed one-third, enough to stall the chain simply by going offline. They all run Hyperliquid's closed-source node software, the only implementation, released as binaries signed with Hyper Foundation keys.
BTC, ETH, and SOL enter through an independent protocol, Unit, so the spot you hold on Hyperliquid is Unit tokens (UBTC, UETH, USOL). Unit holds $940 million across 14 assets, $690 million of which is in BTC, ETH, and SOL. According to its documentation, three guardians (Unit, Hyperliquid, Infinite Field) control these vaults, and every withdrawal requires 2 of the 3 signatures—without them, redemption is impossible. After the Binance incident, Unit added delays to all deposits and withdrawals.
After the JELLY squeeze in March 2025 exposed Hyperliquid's own liquidity vault, validators voted to delist the market and settle at a price of their choosing. The Arbitrum bridge has been emergency-paused multiple times, most recently in November 2025; each time it reopened within hours, with no bridge fund losses.
My assessment: the safety of most USDC depends solely on Coinbase custody and the Hyperliquid chain, with Circle freezes as the backstop; the chain itself relies on a validator set clearly weighted toward the Hyper Foundation, running closed-source software.
Aster: Custody with on-chain speed limits
Aster holds about $500 million across vaults on BNB Chain, Ethereum, Arbitrum, and Solana, $100 million of which is USDF. That's before counting $1 billion in ASTER and APX in the BNB Chain vault.
Vault funds are pooled, and who owns how much is recorded on Aster Chain; that chain is currently run only by Aster's own validators.
Every EVM vault withdrawal must be initiated by an Aster operational address and co-signed by two Aster keys. You also sign the withdrawal request, but the verified code only compares your signature against the address provided by Aster's keys, so it only proves you initiated the request—it does not protect the pool.
This is the same category of custody risk as Binance: if Aster's keys or backend are compromised, funds can leave the vault. Each EVM vault has an hourly withdrawal cap: $30 million on BNB Chain and Ethereum, $20 million on Arbitrum; once the cap is hit, the vault pauses.
Users also cannot withdraw on their own without going through Aster. The verified EVM vault contract code contains no function for users to initiate withdrawals themselves.
EVM vault upgrades go through a 4/7 multisig and a 6-hour timelock, but the same multisig can instantly change roles and caps; the Solana program is upgraded instantly by a 3/6 multisig.
Caps were triggered multiple times in December 2024, September 2025, and July 2026, with withdrawals exceeding limits and pausing the vault; the multisig raised the caps afterward each time. In September 2025, a mispricing in the XPL market led to trader liquidations, and Aster paid compensation within hours. In October 2025, DefiLlama delisted Aster's perpetual trading volume because it closely resembled Binance's, and Aster could not produce order-level data. Its predecessor ApolloX lost $2 million of its own tokens to an exploit in 2022.
My assessment: this is still essentially a custody model. The money sits in Aster's vaults, withdrawals must be initiated by the Aster operator and co-signed by two Aster keys; your own signature only proves "I requested a withdrawal," it does not protect the pool. The on-chain hourly caps do slow down a drain, but they are only a speed limit, not a permission limit. And the same 4/7 multisig that sets the caps can also instantly change roles and raise them—the so-called brake can be released at any time. Ultimately, whether the money is safe depends on how Aster safeguards those two keys and whether its backend systems hold up.
Variational: The least guarded contract design
Variational has $290 million USDC on Arbitrum: $223 million spread across 208,000 user settlement pools, with trader and market-maker margin commingled; $54 million in market-maker OLP addresses; $12 million in the treasury.
You can deposit into pools yourself, but not withdraw. Every withdrawal is initiated by one of 17 operational keys, 16 of which are ordinary wallets. Any single one can move USDC out of any pool, with no amount cap in the contract.
One function sends USDC directly to the calling key's wallet, and a single transaction can process multiple pools in a batch; in theory, a single key could empty all pools in minutes. The admin is also an ordinary wallet with instant effect: adding keys, switching a pool to another oracle contract, pausing all withdrawals. The documentation describes the OLP vault as a smart contract, but the listed address has no code.
Variational uses Fireblocks for contract deployment and oracle signing, but which keys are covered and what the approval rules are is not public. If there are multisig thresholds or limits behind those keys, they are not visible on-chain.
Users likewise have no way to exit without relying on the team. Once the backend stalls, withdrawal requests go unprocessed; the admin can also simply pause the oracle, stalling the withdrawal process. In both cases, the contract has no fallback withdrawal entry point or forced exit function, and users cannot bypass the team to get their money out.
Historically, there have been no reported hacks, thefts, or withdrawal pauses. After the Binance incident, Variational said user funds are safe in on-chain contracts and can be withdrawn at any time; both statements are true, as long as the backend keeps signing. External backstops are Circle freezing USDC or Arbitrum's security committee stepping in, and both have acted after the Binance incident.
My assessment: looking only at verified contracts, Variational has the largest exposure. Any one of the 17 operational keys can move USDC out of any user pool, with no amount cap in the contract; a single key compromise is enough to empty the pools. The only thing in the contract that can intervene is the admin pausing the oracle or removing keys, but the admin is itself a single wallet with instant effect, no timelock, and no multisig threshold. That is more after-the-fact damage control than a proactive defense. Like Aster, the safety of the money ultimately depends on those keys and the backend.
Side-by-side comparison of the four
Of the four, only Lighter lets you exit without relying on the team.
Which is safest?
Based on the three questions above, and looking only at what is verifiable on-chain:
Lighter is the safest by design. Breaching its servers is not enough to withdraw your balance; if the team disappears, you can still use forced withdrawals and desert mode. But the operator also sets prices, and bad prices can still liquidate you. Three votes of the 3/5 multisig plus four votes of the 4/7 security committee—seven votes in total—are enough to change code without notice; every upgrade to date has taken this path.
Hyperliquid sits in the middle. 86% of the USDC is in a single-key Coinbase wallet, and both Coinbase's key and the Hyperliquid chain can move it, with Circle freezes as the backstop.
Aster is in the same tier as Hyperliquid. EVM withdrawals are initiated by the Aster operator and co-signed by two Aster keys. The hourly cap can slow a drain, but the 4/7 multisig can raise the cap at any time.
Variational has the least guarded contract design. Any of the 17 operational keys can move USDC out of any
