What Happens When a DeFi Protocol Gets Hacked

A hack does not end when the funds leave the contract. It starts a sequence of halts, freezes, negotiations and votes, and for most depositors the only step that decides whether they are repaid is the last one. DeFi lost $1.3 billion to exploits in the first half of 2026, according to CertiK, Forbes and TRM Labs data compiled by crypto.news. What happened to depositors afterwards ranged from full recovery to nothing at all, and the code was rarely what made the difference.

Three things did. The first is who holds emergency powers. The second is whether the stolen assets end up somewhere a validator set, a security council or an issuer can freeze. The third is whether a fund exists that is large enough to fill the hole. All three can be checked before a deposit is made. Few depositors check them.

The halt. The first move is to stop everything that still works. When Payy's Ethereum bridge was drained at 4:21 a.m. UTC on September 24, 2026, the network froze deposits, withdrawals, transfers and linked card payments, CryptoSlate reported. No restoration timeline was given. A halt limits further damage. It cannot reverse what has already left, and it is no defense when the privileged keys are the problem: the malicious Payy batch was reportedly sent with the network's own prover key and signed by its own validator key.

The freeze. Whether stolen funds can be caught depends on where they land. Arbitrum's Security Council used a 9-of-12 supermajority to freeze 30,766 ETH belonging to the KelpDAO exploiter on April 21, 2026. Stellar's Tier-1 validators froze about 48 million XLM (roughly $7.2 million) of YieldBlox DAO's $10.97 million loss before it could leave the chain. Attackers know how this works. The Term Finance attacker converted stolen USDC into DAI specifically to avoid a freeze by a centralized issuer.

The offer. Next comes an on-chain message to the attacker's address that offers a share of the loot in exchange for returning the rest. The YieldBlox Security Council, led by Script3, offered 10% and a promise of no legal action. Aquifer offered to let the attacker keep 20% if the rest came back by 14:00 UTC on September 3, 2026. The deadline passed in silence, and the funds were converted into about $2.47 million in USDC and bridged away, according to Shattered.io.

The post-mortem. The post-mortem establishes the cause, and the cause is increasingly not a code bug. One report finds that 67.6% of 2026 incidents fell outside the scope of a standard smart-contract audit, and that those incidents caused 94.4% of losses.

The vote. Last comes the decision on who absorbs the loss. The options are a treasury or protection fund, a shared recovery pool, a deficit spread across depositors, or nobody. In a DAO this goes through the usual forum and token-vote process, which takes weeks, not hours.

KelpDAO. On April 18 an attacker forged LayerZero packets and minted 116,500 unbacked rsETH, which was 18% of the circulating supply. The attacker then posted 89,567 of those tokens as collateral on Aave, Compound and Euler and borrowed about $236 million. The damage therefore reached depositors in lending markets who had never touched Kelp. Recovery became a governance matter. Arbitrum DAO opened a Snapshot temperature check on releasing the frozen ETH to a 2-of-3 Safe controlled by Aave, Kelp DAO and Certora signers. In the first hour, 16.9 million ARB voted in favor and none against. Plaintiffs holding terrorism-related judgments against North Korea then challenged the release in court. A federal court ruled for Aave and permitted the transfer, but Aave still cannot move or sell the funds without further court approval. The proposal also carried an uncapped indemnification clause for the Arbitrum Foundation and Security Council. Industry pledges and loans to the resulting "DeFi United" pool exceeded $311 million, including 25,000 ETH from Aave DAO that was still pending a vote at the time of reporting.

Liquid Network. Strictly speaking this is a Bitcoin sidechain, not a DeFi protocol, but its response followed the same script. About 4,000 of its roughly 4,200 BTC left on September 6 through a flaw in the authorization logic that sits above its 11-of-15 multisig. The attacker called itself a white hat and returned about 3,400 BTC the next day. It kept 598.5 BTC and demanded a 10% bounty, which Blockstream refused and described as a ransom. Block production resumed on September 10.

Solv Protocol. Solv lost $2.73 million to a reentrancy bug and recovered none of it. It had launched without a formal audit or a bug bounty program.

The strongest objection is that recovery happens more often than this account suggests. Liquid got back about 85% of its bitcoin. FOOMCASH recovered 81.4% of a $2.26 million loss after a white hat and the security firm Decurity intervened, and it paid them $420,000 between them. These cases are real, and they show that frozen or negotiated returns can do most of the work.

They do not add up to a rate, however. None of the sources for this page gives an industry-wide share of hacked funds recovered, and the successes are the cases that get written up. Aquifer and Solv recovered nothing. The attackers most active in 2026 were Lazarus-linked, tied to at least $575 million (44%) of first-half losses, and they do not answer bounty messages.

A governance takeover is also a different event from a hack, even though the losses look alike. Term Finance lost $8.5 million after an attacker used its own voting process to reset a seven-day timelock to zero. No bug was exploited, and no negotiation, bounty or recovery vote was reported afterwards.

The basics of wallet hygiene are covered in DeFi security for beginners. The questions here concern the protocol itself.

The first is whether a safety module, treasury reserve or protection fund exists, and how large it is compared with deposits. The sources for this page do not give the size of any DeFi protocol's module, so that figure has to be read from the protocol's own documentation, and if it cannot be found, that is itself the answer. Outside insurance will not make up for a missing fund. By one estimate, on-chain insurance capital totaled $121 million against $93.9 billion of DeFi TVL, and Nexus Mutual had paid about $18 million in claims over seven years. Its founder, Hugh Karp, puts coverage at under 2% of TVL.

The second is who can pause the protocol and who can freeze the assets it holds, because the same keys that can stop an attack can also be stolen.

The third is exposure through collateral. KelpDAO showed that one protocol's unbacked token can put a hole in another protocol's lending market, so the relevant question is what a market accepts as collateral, not only who built the market. For how these protocol types connect, see the wider guide to how DeFi protocols work.

If no fund exists, depositors should assume that after a hack they will be repaid either by a vote or not at all.