$292M Exploit, $14B in DeFi Exits: What Allocators Can Learn From April 18

Want to skip ahead and read the full report? Read it here.

On April 18, 2026, a forged LayerZero message released 116,500 rsETH (approximately $292M, or 18% of circulating supply) from Kelp DAO’s Ethereum-side adapter. No corresponding ETH was locked on any source chain.

Within minutes, the attacker converted the unbacked tokens into ~$236M of WETH and wstETH borrowings across Aave V3/V4, Compound V3, and Euler. Over the following 72 hours, approximately $14B of DeFi total value locked exited across the top 20 chains, most of it from protocols with zero direct exposure to rsETH or Kelp.

Every smart contract in the event performed exactly as designed. Every one was audited. The damage differential, from $292M at Kelp to roughly $1M at Morpho, was a function of architectural choice upstream of code quality.

That finding is the core of this handbook.

Four protocols, one forged message, four different outcomes

Kelp DAO was the origin: the bridge operated on a 1/1 DVN configuration, meaning a single signer needed to approve the forged message with no quorum redundancy. When the signer approved it, $292M of unbacked rsETH entered circulation.

Aave V3/V4 took the largest downstream damage: $123–236M in direct bad debt across five chains, plus $6–10B in sentiment- driven TVL outflows over 24 to 72 hours. Aave’s pooled-whitelist architecture means that when any whitelisted asset fails, the loss is mutualized across all depositors.

Morpho V1/V2 took about $1M of direct bad debt across 2 of its ~500 markets. Morpho’s isolated-market architecture contained the blast radius to just the affected pairs.

Lido’s earnETH aggregator took indirect damage through a 9.1% levered rsETH/ETH loop on Aave, about $21.6M stressed. It paused deposits on April 19.

Four protocols received the same attack. The damage spread three orders of magnitude because of architecture, not code.

Why current diligence frameworks missed this

Allocator diligence frameworks calibrated to 2022–2023 smart- contract threats are structurally mismatched against 2025–2026 reality.

Per TRM Labs, code exploits accounted for only 12.1% of 2025 stolen value ($350M across 52 incidents). Infrastructure-layer and human-factor attacks drove the majority. Per Chainalysis, DPRK-linked actors accounted for approximately 76% of 2025 service-compromise value, at least $2.02B. The April 2026 cluster alone (Resolv Labs, Drift, Rhea Finance, Kelp) shows more than $575M drained in 18 days across two structurally different vectors.

Audits still matter. They are necessary but no longer sufficient.

The handbook

We built a seven-stage allocator decision sequence, each stage anchored to a specific question allocators should answer before capital goes into a DeFi protocol.

1. Mandate. Four questions that shape every downstream choice: liquidity horizon, composability dependency, regulatory posture, diligence capacity.
2. Pattern. Five observable patterns of institutional DeFi participation, each with a different blast-radius profile: pooled lending, isolated markets, aggregator vaults, covered vaults, and permissioned venues.
3. Diligence. A seven-layer risk stack that organizes counterparty evaluation into testable categories. Layer 4 (cross-chain messaging) and Layer 6 (composability amplification) are now first-order diligence targets, both drove April 18 outcomes and are commonly underweighted.
4. Position construction. Size for blast radius, not notional. Decompose structural from sentiment exposure. Price the coverage gap explicitly.
5. Monitoring. Daily, weekly, monthly, and quarterly signals, each mapped to the risk layer it informs.
6. Incident response. A pre-defined protocol for the 46-minute origination window. Allocators who rely on real-time human judgment in that window lose money in predictable ways.
7. Context. The threat surface has shifted. April 18 is consequential for crypto-native allocators. For mainstream TradFi it reinforces existing caution.

What this handbook is not

This is not an allocation recommendation. It does not tell you which protocols to deploy into. It structures the questions you should answer before deploying, and the signals to monitor once you do.