Perp DEX Security: Who Can Move Your Funds?
Compare how four perp DEXs hold and release user funds, what a compromised team could do, and whether traders can exit without operator approval.

Binance Was Drained. Could Your Perp DEX Be Next?
A crypto exchange doesn't have to lose its private keys for users to lose access to their money. If an attacker compromises the systems that prepare transactions - or the systems that instruct signers - withdrawals can still go through. That's the uncomfortable question raised by the reported Binance loss of $6.3 billion from hot wallets on October 29: if an exchange's infrastructure can authorize a transaction, what does keeping funds onchain actually protect you from?
Perpetual DEXs are often pitched as the alternative to centralized custody. But "decentralized" doesn't automatically tell you where funds sit, who can move them, or whether you can withdraw if the team disappears. Those details vary a lot by protocol. This review compares Lighter, Hyperliquid, Aster and Variational using three practical questions: where the funds are held, whose authorization can move them, and whether users have an independent exit. This is a contract-level risk check based on published information - not an audit, and not a prediction that any of these systems will be exploited next.
Why a compromised backend matters
The reported Binance incident is a good reminder that "custody risk" is broader than private-key theft. According to Binance's initial account, an attacker accessed an internal service after compromising a DevOps engineer's laptop, then initiated withdrawals from hot wallets. Binance said there was no indication that private keys had leaked, and it promised to cover the losses. Some funds were frozen, but the exchange hasn't published a full post-mortem.
On Ethereum, TRM reported that the attacker's transaction received approval for an EIP-7702 delegation - meaning an attacker-controlled contract gained control over a hot wallet. Researchers suggested the signing system may have relied on blocking known-bad actions rather than allowing only known-good ones. Binance hasn't confirmed that account, and it hasn't explained reported losses on other networks. So treat the technical story as an informed report, not the final, complete explanation.
A month earlier, Bitget reported a loss of more than $350 million after attackers compromised a backend system that supplied transaction data to signers. In both cases, the same distinction matters: who holds a key versus what the systems around that key will authorize. A signature can be produced by a well-secured key-management setup and still approve a harmful transaction if the approval process is compromised - or too permissive.
The three questions that matter
For a perp DEX, "funds are onchain" isn't enough. You want to know:
- Where does the money sit? In a contract with enforceable rules, a wallet controlled by an operator, or some mix?
- Who can move it? Does a user authorize each withdrawal, does the contract enforce it, or can an operator key withdraw on the user's behalf?
- Can you leave without the team? If the backend stops, an admin pauses withdrawals, or the team is compromised, do you have a forced exit or a real escape hatch?
These questions separate things you can see in contract guarantees from protections outsiders can't verify. A project may use MPC, hardware security modules, signing policies, or monitoring. Those can help - but unless the limits are enforced in the contracts, users can't independently confirm what they actually prevent.
What this review does - and does not - show
The comparisons below are based on onchain balances, verified contract code, project documentation, and public incident reporting, including L2BEAT's risk analysis. Balances and contract setups can change. This is a focused scenario check: what might happen if a team's systems or keys were compromised, and what could users do next?
It's not a full audit, a bug hunt, or a claim that any protocol is currently compromised. And a contract-level weakness doesn't prove an attacker can exploit it easily. The point is narrower: what does the code permit, and how much safety depends on trusting the operator?
How four perp DEXs control withdrawals
The protocols below take different approaches. Lighter leans toward user ownership and proof-checked batches, while the other three rely more directly on operator-controlled systems or keys for withdrawals. That doesn't mean there's a single clean ranking of "most safe." Market-price controls, chain governance, stablecoin freezes, and upgrade powers all affect the risk too.
Lighter: proof-based withdrawals and an escape hatch
Approximately $1 billion is in Lighter's Ethereum contract, mainly USDC and LIT. In normal operation, the contract releases funds in batches that pass a zero-knowledge proof check. A standard withdrawal can go only to the wallet tied to the account at creation. Faster withdrawals and transfers between accounts require the user's wallet signature.
The practical effect is big: even if you can compromise Lighter's sequencer, prover, or validator systems, that alone isn't enough to pull out user balances. Users also have a forced-withdrawal route. If Lighter doesn't process a request within 14 days, anyone can activate "desert mode." Upgrades stop, and users can exit using a proof of their balance derived from data already published on Ethereum.
This exit mechanism doesn't eliminate every trust assumption. Lighter's operator provides index and mark prices, and the proof system doesn't verify those prices against oracle signatures. A bad price could still trigger unfair liquidations without directly stealing funds. Two guardian keys can pause processing or withdrawals for limited windows - up to 24 or 72 hours - though neither can disable desert mode.
Upgrades are another real risk area. A 3-of-5 multisig can initiate upgrades with a 21-day public notice. A separate 4-of-7 security council can co-sign to remove that delay. The source review notes that upgrades have used the fast-track route, including repeated verifier changes in October 2026. Lighter's design reduces operator withdrawal power, but users still depend on price inputs and on governance not changing the rules in ways that surprise you.
Hyperliquid: substantial third-party custody and chain dependence
About $6.5 billion in USDC backs Hyperliquid, split across three locations. Roughly 86% is in a Coinbase-named wallet on HyperEVM, around 10% is in Circle's upgradeable CoreDepositWallet contract, and the remaining 4.4% is in an older bridge on Arbitrum.
The largest chunk sits in a plain wallet controlled by a single key. The chain can also move USDC into and out of that wallet. Coinbase has staked HYPE for its role, but the reported stake is under 1% of the USDC balance, and it can be slashed only under a specific failure to pay the protocol its share of yield. That stake isn't collateral that matches the value of the funds.
Circle's contract has a single-address admin and can be upgraded. The Arbitrum bridge uses three-of-four validator signatures, with a 200-second window where a withdrawal can be stopped. Circle can freeze native USDC - useful as a theft backstop - but frozen funds aren't automatically available to users right away.
There are other dependencies beyond custody. Hyperliquid has 27 active validators, with substantial stake concentrated among Hyper Foundation validators. They run closed-source node software distributed as signed binaries. The chain has also shown validators can make consequential market decisions: in March 2025, validators voted to delist JELLY and settle the market at a chosen price after a squeeze exposed the protocol's liquidity vault. That's evidence of governance power, not proof of a fund theft.
Spot BTC, ETH, and SOL are represented by Unit-issued tokens backed by assets held by Unit. Unit's documentation describes a three-guardian arrangement, with two-of-three signatures required for withdrawals. So redemption depends on those guardians. Hyperliquid's setup combines institutional custody, protocol-controlled movement, and third-party freeze powers; it's not a "single user-controlled wallet" model.
Aster: operator-authorized withdrawals with adjustable limits
Aster's vaults hold roughly $500 million across BNB Chain, Ethereum, Arbitrum, and Solana, plus a substantial balance of Aster's own tokens. On EVM chains, user balances are pooled in vaults while ownership is tracked on Aster Chain, which currently relies on Aster's validators.
An EVM withdrawal requires an Aster operator address to send the transaction, plus two Aster keys to sign off. The user also signs a withdrawal request, but the verified code checks that signature against an address supplied by Aster's keys. In other words, the signature confirms the request; it doesn't independently protect the pooled vault from an operator-controlled withdrawal.
Hourly withdrawal caps limit outflow: $30 million on BNB Chain, $30 million on Ethereum, and $20 million on Arbitrum. When a cap is hit, the vault pauses. That can slow a drain, but it's not a hard ceiling on what the team can move over time. A 4-of-7 multisig can change roles and caps immediately, even though ordinary EVM vault upgrades have a six-hour timelock. A separate 3-of-6 multisig can upgrade the Solana program without a timelock.
The caps have triggered pauses in the past, after which the multisig raised them. That suggests the limits are adjustable in practice. Aster's model may give useful speed controls, but users can't withdraw independently from the EVM vaults. If relevant keys or backend systems are compromised, the contract-level protections provide less separation from team custody than Lighter's forced-exit design.
Variational: broad operator withdrawal capability
Variational holds about $290 million in USDC on Arbitrum. Most is spread across roughly 208,000 user settlement pools, with additional funds in a market-maker address and treasury. Users can deposit into their pools, but withdrawals are executed by operator keys.
The contract design gives 17 keys the relevant provider role. Based on the verified code reviewed for this comparison, any one of those keys can withdraw USDC from any user pool, and there's no withdrawal cap in the contract. A batch function can move funds from multiple pools in one transaction. The admin is a single wallet and can change operator permissions or pause withdrawals immediately.
Variational says it uses Fireblocks for contract deployments and oracle signing, but public information doesn't establish which keys are covered or what approval rules apply. Those offchain controls might matter, but they don't change what the contracts allow if an authorized key is misused. The reviewed contracts also don't provide an independent user exit if the backend stops or withdrawals are paused.
There are no reported hacks, thefts, or withdrawal halts in the source material. After the Binance incident, Variational said user funds are safe in onchain contracts and can be withdrawn at any time. Both statements still depend on the backend continuing to sign withdrawals. Circle's ability to freeze USDC and Arbitrum's Security Council intervention are external backstops, not user-controlled exits.
Which perp DEX offers the strongest exit protections?
The most useful comparison isn't "centralized versus decentralized." It's whether a compromised team can move user funds, and whether users can recover those funds without the team.
| Protocol | Where funds sit | Who can move them? | Can users exit without the team? | Main limits or backstops |
|---|---|---|---|---|
| Lighter | About $1B in an Ethereum contract | Proof-verified batches; standard withdrawals return to the account owner's wallet | Yes, through forced withdrawal and desert mode after 14 days | Guardian pauses are time-limited; upgrades can be fast-tracked by a security council |
| Hyperliquid | About $6.5B USDC: mostly in a Coinbase-named wallet, with Circle and Arbitrum bridge balances | Coinbase's key and the chain can move the largest share; bridge uses 3-of-4 signatures | No independent exit described | Circle can freeze USDC; bridge withdrawals have a 200-second stop window |
| Aster | About $500M in vaults across four chains | Operator transaction plus two Aster keys for EVM withdrawals | No independent EVM-vault exit described | Hourly caps can pause withdrawals, but the multisig can change caps and roles |
| Variational | About $290M USDC on Arbitrum | Any one of 17 provider keys can withdraw from user pools | No independent exit described | Admin can pause or remove keys; Circle and Arbitrum provide external intervention |
A practical reading of the comparison
Lighter has the strongest user exit by design. A server compromise on its own shouldn't let an operator withdraw account balances, and the forced-exit path matters if the team stops responding. The trade-off is continued reliance on operator-supplied prices and on upgrade governance. An exit protects custody; it doesn't stop a bad liquidation before you can use it.
Hyperliquid and Aster have different versions of concentrated control. Hyperliquid's biggest USDC pool depends on a named institutional wallet and the protocol's chain. Aster's EVM withdrawals depend directly on its operator and two signing keys, with caps that can be raised. Neither matches Lighter's independent exit model described here. Whether one is "safer" overall depends on what you care about most: custody, governance, chain liveness, or the external institutions involved.
Variational has the broadest onchain operator authority in this review. Any one of its provider keys can withdraw from user pools without a contract-level amount cap. That's what the verified code permits - not proof that a key is compromised. But it does create a clear concentration risk: a single key compromise could affect many pools.
None of this means you should automatically avoid these venues. Perp trading always involves liquidation, funding, oracle, and market risks, regardless of how custody is designed. The point is simpler: "DEX" doesn't answer the custody question. Look at the actual withdrawal path.
How to assess custody risk before trading
You don't need to re-run a full smart-contract audit to ask the right questions. Start with where the money lives, then follow the authority that can move it. A good quick review should cover both the normal path and the failure path.
A five-step review
- Locate the assets. Identify the relevant chain and contract addresses. Check whether balances are in a user-specific contract, a pooled vault, a bridge, or a plain wallet. A product interface isn't evidence of custody structure.
- Trace a withdrawal. Find the function that releases funds. Determine what signatures or roles it needs, whether the user's signature is actually checked against the user's address, and whether one operator can withdraw for many accounts.
- Look for hard limits. A cap only matters if the same authority can't raise or remove it instantly. A pause can limit damage, but it can also trap users. Check who controls both actions.
- Test the team-disappears scenario. Look for a forced withdrawal, escape hatch, delayed exit, or recovery procedure. Note what triggers it and how long users have to wait. If the feature still requires an operator to process the transaction, it isn't truly independent.
- Review upgrade and governance powers. Timelocks only provide notice if administrators can't bypass them. Check multisig thresholds, emergency roles, verifier upgrades, chain validators, and any third-party issuer that can freeze the asset.
Keep the risk proportional to the use
A perp venue is usually for trading, not for storing funds long-term. Keeping only the margin you need reduces exposure if withdrawals stop or custody assumptions fail. That's not a guarantee - fast market moves can still liquidate positions, and moving funds between chains adds its own risks. But it helps you avoid treating a trading balance as if it has the same protection as assets in a wallet you control.
For larger balances, compare the value at risk with the weakest important dependency. If one key can withdraw from every pool, a strong brand or high TVL doesn't remove that authority. If you can only exit after a delay, plan around that delay. If an issuer can freeze the stablecoin, treat it as a third-party recovery mechanism - not a substitute for a user-controlled exit.
Conclusion
A perp DEX's security isn't captured by its name, its TVL, or the fact that its contracts are visible. The relevant questions are concrete: where the funds sit, who can authorize withdrawals, what limits that authority, and whether users can leave if the team is unavailable or compromised.
In this comparison, Lighter stands out for proof-based withdrawals and an independent exit route, while still carrying price and governance risks. Hyperliquid, Aster, and Variational rely more directly on operator, custodian, or validator power, with different safeguards and trade-offs. Before committing meaningful capital, trace the withdrawal path yourself - and keep only the funds you need for the trade there.