Lesson 5 of 7

Smart contract, oracle, bridge and MEV risk

Step 3 of 6

Worked example

Two bridge failures, from official reports unless marked.

Ronin bridge (2022) KelpDAO rsETH bridge (2026)
Meant to Hold assets for the Axie Infinity game's chain; withdrawals needed 5 of 9 validators' signatures Back copies of rsETH (a Kelp token backed by staked ETH) on other chains with rsETH locked on Ethereum
What failed An attacker got 5 keys: four of Sky Mavis's, after an employee was phished, and one of the Axie DAO's, through a Nov 2021 permission never revoked On 18 Apr 2026, 17:35 UTC, the route from Unichain relied on a single verifier. It accepted a forged message and released 116,500 rsETH with nothing burned on the other side
Cost 173,600 ETH + 25.5M USDC, about $624M at the time (DefiLlama, rekt) About $293M (DefiLlama); about $290M (rekt)
Found 23 Mar 2022, noticed 29 Mar: "we didn't have a proper tracking system for monitoring large outflows" Same day

How KelpDAO reached a lending market. Aave's incident report (20 Apr 2026) says the attacker deposited 89,567 of the rsETH on Aave and borrowed 82,650 WETH ($190.86M). The locked balance behind the copies had fallen from 116,723 to 223 rsETH, leaving Aave with bad debt: loans the collateral can no longer cover. Aave froze rsETH from 18:52 UTC, saying the incident did "not stem from a vulnerability in the Aave protocol itself". WETH hit 100% utilisation (all of it lent out), so depositors couldn't withdraw for a time.

Fault is disputed. Press reports say LayerZero, the messaging system, blamed Kelp's single-verifier set-up; Kelp called it LayerZero's default.

Afterwards:

  • Ronin. The post-mortem (27 Apr 2022) says user funds were "in the process of being restored". Repayment, the June 2022 reopening and the FBI's blaming of North Korea are press reports. A second exploit, about $12M, followed an incomplete upgrade on 6 Aug 2024 (DefiLlama).
  • KelpDAO. A coalition, "DeFi United", funded the shortfall; an Aave proposal of 24 Apr 2026 asked to contribute 25,000 ETH. Aave Labs reported the attacker's positions liquidated on 7 May and rsETH withdrawals reopened on 14 May 2026. Kelp's 25 May statement that recovery was complete is known from press reports.

What an outsider could see. A few signers or one verifier guarding hundreds of millions can be found; who held the keys, or a forgotten permission, can't. Aave's depositors never used the bridge: Aave accepting the token as collateral was enough.

Rankings disagree, so check rekt.news against DefiLlama. rekt's #1, the 2020 LuBian theft, shows $14.85B at today's prices; DefiLlama lists $3.5B. By DefiLlama's count the three largest DeFi losses are all bridges, Ronin first.

Checklist: before trusting a protocol, ask

  1. How long has its code run with this much money in it?
  2. Who can upgrade it or change its settings, and how fast?
  3. Where do its prices come from?
  4. Does it rely on a bridge, and how many signers approve a withdrawal?
  5. What backs the tokens it accepts as collateral?
  6. Has it lost money before?