DeFi protocols have hemorrhaged $1.1 billion to exploits over the past 12 months. April 2026 recorded 27 separate hacking incidents — the highest monthly count in crypto history — draining $690 million in a single month. Total value locked across DeFi fell 14%, from $172 billion to $148 billion, ...
"Coding agents are superhuman at finding vulnerabilities, and smart contract security is too asymmetric: defenders need to fix every bug while attackers need just one exploit to steal funds." — Manuel Aráoz, Co-founder, OpenZeppelin
DeFi protocols have hemorrhaged $1.1 billion to exploits over the past 12 months. April 2026 recorded 27 separate hacking incidents — the highest monthly count in crypto history — draining $690 million in a single month. Total value locked across DeFi fell 14%, from $172 billion to $148 billion, as users withdrew more than $20 billion since January.
The structural trigger: AI coding agents can now detect and weaponize smart contract vulnerabilities faster than human auditors can patch them. OpenAI and Paradigm's EVMBench, released in February 2026, showed GPT-5.3-Codex exploiting 72.2% of critical vulnerabilities in sandboxed tests — up from under 20% when the benchmark launched. OpenZeppelin co-founder Manuel Aráoz declared on May 26 that he considers "all of DeFi unsafe" and advised associates to exit positions entirely. CertiK CEO Ronghui Gu stated that in April alone, "there were only three days without hacks."
The attacker-defender asymmetry has widened to a point where the sector's $2.69 billion audit market may be structurally insufficient to protect $148 billion in locked capital.
Through the first five months of 2026, DeFi protocols have lost over $1.1 billion to hacks and exploits, according to DefiLlama data. The damage concentration is severe: two attacks alone — Kelp DAO ($292 million) and Drift Protocol ($285 million) — account for more than half of all losses.
Monthly breakdown:
| Month | Estimated Losses | Notable Incidents | |-------|-----------------|-------------------| | January | ~$27M | Step Finance oracle overflow | | February | ~$40M | Multiple smaller exploits | | March | ~$70M | Resolv Labs stablecoin mint exploit, TrueBit | | April | ~$690M | Kelp DAO ($292M), Drift ($285M), 25+ others | | May (through 28th) | ~$50M+ | 25 recorded incidents, smaller individual losses |
April 2026 set a record as the most-hacked month in cryptocurrency history by number of incidents, according to Crowdfund Insider, with approximately 27-30 separate exploit events. May has already logged 25 additional incidents, though individual losses have been smaller.
The attack vector has shifted. According to TRM Labs and CertiK analyses, the primary targets are no longer core smart contract logic but rather bridge layers, oracle systems, signing infrastructure, and multisig key holders — components that are harder to audit via traditional code review.
In February 2026, OpenAI and Paradigm jointly released EVMBench, an open-source benchmark measuring AI agents' ability to detect, patch, and exploit smart contract vulnerabilities across 120 curated scenarios drawn from 40 professional audits.
The results quantified what security practitioners had feared:
| Model | Exploit Mode Success Rate | |-------|--------------------------| | GPT-5.3-Codex | 72.2% | | GPT-5 | 31.9% | | Earlier models (pre-2026) | <20% |
In exploit mode, GPT-5.3-Codex successfully generated end-to-end fund-draining attacks against 72.2% of critical, high-severity vulnerabilities — a more than threefold improvement from earlier model baselines.
Separately, research from Anthropic's red team demonstrated that Claude Opus 4.5 and GPT-5 developed working exploits collectively targeting $4.6 million in vulnerable contracts deployed after their training data cutoffs. Both models uncovered two novel zero-day vulnerabilities in recently deployed contracts, producing functional exploits at an API cost of $3,476, according to Anthropic's published findings.
The cost asymmetry is stark. According to Nadcab Labs, AI-powered vulnerability scanners can audit legacy smart contracts for as little as $1.22 per scan. Defenders, by contrast, pay $50,000 to $100,000 for a mid-complexity DeFi protocol audit, with formal verification adding $20,000 to $50,000 on top, according to Zealynx Security's 2026 pricing data.
OpenZeppelin audited the EVMBench methodology itself and identified flaws — at least four issues labeled high severity were not exploitable in practice, and training data contamination questions remain. But the directional signal is unambiguous: AI exploit capabilities are improving faster than defensive tooling.
Attackers exploited Kelp DAO's rsETH cross-chain bridge built on LayerZero infrastructure. The method was infrastructure-level, not a smart contract bug:
Approximately $75 million was frozen on Arbitrum. The remaining proceeds were laundered through THORChain, converting stolen ETH to Bitcoin. The entire drain took 46 minutes, according to CoinCentral's reconstruction.
The Drift breach on Solana involved months of social engineering, not a code exploit:
Drift outlined a recovery plan for users on May 5, according to CoinDesk. The breach is the second-largest exploit in Solana history.
Neither attack exploited a smart contract vulnerability in the traditional sense. Both targeted operational infrastructure — RPC nodes, signing authority, oracle feeds — layers that fall outside standard audit scope.
According to TRM Labs data published in April 2026, North Korea-linked actors accounted for 76% of all crypto hack value in 2026 through April, stealing approximately $577 million via the Drift and Kelp DAO attacks alone.
The escalation trajectory, per TRM Labs:
| Year | North Korea Share of Global Crypto Hack Losses | |------|-----------------------------------------------| | 2020 | <10% | | 2021 | <10% | | 2022 | 22% | | 2023 | 37% | | 2024 | 39% | | 2025 | 64% | | 2026 (through April) | 76% |
Cumulative attributed North Korean crypto theft now exceeds $6 billion since 2017, according to TRM Labs. The Kelp DAO breach was attributed to TraderTraitor, a well-documented Lazarus-affiliated group. The Drift attack was attributed to a separate North Korean subgroup, suggesting multiple state-sponsored teams operating concurrently.
THORChain has emerged as the consistent laundering bridge across North Korea's largest operations, processing the majority of proceeds from both the 2025 Bybit breach ($1.46 billion) and the 2026 Kelp DAO hack. No THORChain operator has frozen or rejected these transfers.
The smart contract security audit market reached an estimated $2.69 billion in 2026, up from $2.14 billion in 2024, according to industry data compiled by CoinLaw. Protocols with meaningful TVL now spend $150,000 to $500,000 annually on security, with leading protocols allocating 15-20% of development budgets to security-as-a-service.
However, the audit model faces a structural mismatch:
Attack surface expansion: The shift from monolithic smart contracts to multi-chain bridges, oracle networks, multisig infrastructure, and off-chain components means the audit perimeter has expanded far beyond what a code review covers.
Speed disparity: An AI agent can scan thousands of contracts in hours. A thorough manual audit of a mid-complexity protocol takes 4-8 weeks and costs $50,000-$100,000. By the time an audit concludes, the codebase may have changed.
Insurance gap: Nexus Mutual, the largest DeFi insurance protocol, has protected over $6 billion in digital assets since 2019 and generated $5.7 million in cover fees in 2025. These figures are negligible against $1.1 billion in annual losses. The insurance pool cannot absorb systemic risk at current scale.
OpenZeppelin's current CEO Demian Brener issued a counterpoint to Aráoz's warning, stating that the firm advocates "continuous, AI-augmented security rather than retreat from DeFi." The implication: defenders must adopt the same AI tools as attackers to maintain parity.
CertiK responded by launching an "AI Skill Scanner" product — described as antivirus software for AI agents — designed to scan AI agents for malicious behavior before they access sensitive data or accounts. Whether defensive AI can close the gap with offensive AI remains unproven.
DeFi's total value locked declined from approximately $172 billion in mid-April to roughly $148 billion by late May — a 14% contraction representing over $20 billion in capital withdrawal since the start of 2026, according to DefiLlama data cited by CoinDesk.
The causal chain: high-profile exploits erode user confidence, triggering withdrawals that reduce liquidity, which in turn increases slippage and reduces yield, prompting further withdrawals. This reflexive dynamic is observable in TVL data following each major incident.
Step Finance, a Solana-based portfolio dashboard, shut down operations entirely following its January exploit. The protocol's closure illustrates the terminal risk for smaller DeFi projects facing security costs that exceed their revenue.
The economic question for DeFi, consistent with the sector's broader sustainability challenge, is whether fee revenue justifies security expenditure. At $148 billion TVL generating approximately $10.6 billion in annualized protocol revenues (per webthreepedia's foundational analysis), the sector spent roughly $2.69 billion on audits and security — approximately 25% of revenue — while still losing $1.1 billion annually to exploits. The net security deficit effectively functions as an unpriced tax on DeFi users.
DeFi's security model was designed for an era when attackers were human, audits were comprehensive, and smart contracts were the primary attack surface. None of those assumptions hold in 2026.
The sector now faces a cost-of-capital problem: if the probability-adjusted loss rate from exploits exceeds the yield premium DeFi offers over traditional alternatives, rational capital will migrate. The $20 billion TVL decline since January suggests this migration is underway.
The path forward likely bifurcates. Protocols with sufficient revenue to fund continuous AI-augmented security, formal verification, and operational hardening may survive the selection pressure. Smaller protocols with limited security budgets face existential risk — as Step Finance's closure demonstrated.
The structural question is whether DeFi can internalize its security costs and still offer competitive returns. The data, so far, is inconclusive.