DeFi protocols lost at least $138.35 million across 15 documented exploits in Q1 2026, according to tracker data aggregated by Cryptopolitan and CryptoAdventure, exceeding Q1 2025 losses of $106.8 million by 29.5%. The quarter's damage was concentrated in four incidents: Step Finance ($27.3 milli...
"Claude Opus 4.6 wrote vulnerable code, leading to a smart contract exploit with a $1.78 million loss. Is this the first hack of vibe-coded Solidity code?" — Pashov, Smart Contract Security Auditor
DeFi protocols lost at least $138.35 million across 15 documented exploits in Q1 2026, according to tracker data aggregated by Cryptopolitan and CryptoAdventure, exceeding Q1 2025 losses of $106.8 million by 29.5%. The quarter's damage was concentrated in four incidents: Step Finance ($27.3 million), Truebit ($26.2 million), Resolv ($25 million), and SwapNet ($13.4 million). Private key compromise, not smart contract logic errors, was the dominant attack vector.
A new variable entered the threat model. In February, the Moonwell lending protocol lost $1.78 million after an oracle misconfiguration linked to AI-generated code — what security researchers have termed "vibe coding." Pull request 578 on the Moonwell repository was co-authored by Claude Opus 4.6, an AI coding assistant, and contained a pricing formula error that went undetected through governance and deployment. The incident coincided with a Tenzai study that found 69 vulnerabilities across 15 applications built with five major AI coding agents. OpenAI responded by launching EVMbench, a smart contract security benchmark, with $10 million in API credits for defensive research. The quarter illustrates a structural tension: DeFi's $105 billion in TVL is protected by audit infrastructure that demonstrably fails to catch deployed vulnerabilities, while a new generation of AI-assisted code introduces error patterns that existing review processes were not designed to detect.
PeckShield data records $112.53 million in crypto hack losses across January and February 2026 alone. Adding March incidents — including the $25 million Resolv exploit on March 22, the $128 million Balancer exploit aftermath, and additional smaller incidents — the Q1 DeFi-specific tally exceeds $138 million.
Monthly breakdown (protocol hacks only):
| Month | Incidents | Total Losses | Largest Single Incident | |-------|-----------|-------------|------------------------| | January 2026 | 16 | $86.01M | Step Finance ($27.3M) | | February 2026 | 15 | $26.52M | YieldBlox DAO ($10M) | | March 2026 (through 3/23) | ~5 | $25M+ | Resolv ($25M) | | Q1 Total | ~36 | $138.35M+ | Step Finance ($27.3M) |
For context, January 2026's combined crypto theft — including social engineering and phishing — reached approximately $370 million, according to CertiK. A single phishing victim lost $284 million, representing over 80% of the month's social engineering total. Protocol hacks accounted for just 23% of January's total theft.
February saw a 69.2% decline from January and a 98.2% year-on-year decrease compared with February 2025, when the $1.4 billion Bybit exploit dominated the figures.
Attack vector distribution across Q1 2026 incidents:
The Solana-based portfolio tracker suffered the quarter's largest single loss when attackers gained access to executive team devices, compromising private keys controlling treasury wallets. The attackers unstaked 261,854 SOL before transferring funds — a pattern consistent with key compromise rather than protocol-level exploitation. Step Finance announced permanent closure on February 24. The STEP token lost 96% of its value. Affiliate projects SolanaFloor and Remora Markets also shut down.
Root cause: Device compromise leading to private key exfiltration. No smart contract vulnerability was involved.
The first DeFi hack of 2026 targeted a five-year-old closed-source smart contract within the Truebit ecosystem. The attacker exploited an integer overflow vulnerability in a purchase contract, setting the token purchase price to zero and extracting 8,535 ETH ($26.44 million) at near-zero cost. The TRU token collapsed 99.9% within 24 hours.
Root cause: Overflow vulnerability in a legacy closed-source contract that had never been updated or re-audited since deployment.
An attacker gained access to Resolv's AWS Key Management Service, where a privileged signing key was stored. Using the protocol's own permissions, the attacker minted 80 million unbacked USR tokens — depositing just 100,000 USDC and receiving 50 million USR in return. The stablecoin crashed from $1.00 to $0.025 on Curve Finance within 17 minutes. According to Chainalysis, the exploit stemmed from structural design failures: single-key controlled privileged account, no oracle or amount checks, and no maximum mint limits.
Root cause: Compromised AWS KMS key combined with absence of minting guardrails.
Attackers leveraged an arbitrary call flaw in SwapNet's DEX aggregator contracts, draining funds from users who had granted infinite token approvals. The confirmed losses of $13.43 million hit 20 users across Ethereum, Arbitrum, Base, and Binance Smart Chain. One user lost $13.34 million. SwapNet paused contracts 45 minutes after the initial exploit on Base, but 13 additional users were affected on other chains during that window.
Root cause: Insufficient input validation on call data, enabling arbitrary token transfers via existing approvals.
On February 15, 2026, Moonwell DAO executed proposal MIP-X43, enabling Chainlink OEV wrapper contracts across its Base and Optimism markets. One oracle was misconfigured: instead of deriving cbETH's USD price by multiplying the cbETH/ETH feed by ETH/USD, it transmitted only the raw cbETH/ETH ratio. The oracle reported cbETH at $1.12 instead of approximately $2,200.
Liquidation bots immediately attacked collateralized cbETH positions, repaying approximately $1 of debt and receiving 1,096.317 cbETH in return. Total bad debt: $1.78 million.
Security auditor Pashov identified that pull request 578, which introduced the faulty configuration, was co-authored by Claude Opus 4.6. The observation ignited an industry debate, though Pashov noted that "behind the AI is a person who checks the finished work, and possibly a security auditor."
The Moonwell incident did not occur in isolation. A December 2025 study by security firm Tenzai tested five AI coding agents — Claude Code, Cursor, Windsurf, Replit, and Devin — using identical prompts to build 15 applications. Results:
A separate March 2026 study by Help Net Security confirmed that AI coding agents "keep repeating decade-old security mistakes," with authorization and access control errors dominating the findings.
The pattern is consistent: AI coding tools generate syntactically correct, functionally working code that systematically omits context-dependent security controls — exactly the type of checks that prevent oracle misconfigurations, unauthorized minting, and privilege escalation.
The Q1 data reinforces a structural problem in DeFi security infrastructure. According to Coinlaw's 2026 audit statistics report, the median time between a DeFi protocol passing an audit and getting exploited is 47 days. Between 2020 and 2025, over $4.2 billion was drained from audited protocols.
The audit market itself reflects this tension. A mid-complexity DeFi protocol's pre-launch security budget in 2026 ranges from $60,000 to $120,000, according to Sherlock's pricing benchmark. Enterprise-grade multi-chain systems can exceed $250,000. Yet audit scope typically covers a snapshot of code at a fixed point in time. Post-audit changes — including governance proposals like Moonwell's MIP-X43 — frequently bypass the original audit perimeter.
Attack vectors in Q1 2026 illustrate the gap:
The common thread: the exploited surface is increasingly outside the boundaries of what traditional smart contract audits cover.
On February 18, 2026 — three days after the Moonwell exploit — OpenAI and Paradigm launched EVMbench, a benchmark evaluating AI agents' ability to detect, patch, and exploit smart contract vulnerabilities. The benchmark draws from 120 vulnerabilities identified across 40 prior audits, plus vulnerability scenarios from Paradigm's Tempo blockchain.
Performance results:
| Task | GPT-5.3-Codex | GPT-5 | |------|--------------|-------| | Exploit mode | 72.2% | 31.9% | | Detection mode | Lower | Lower | | Patching mode | Lower | Lower |
The exploit score of 72.2% means GPT-5.3-Codex could successfully drain funds from vulnerable contracts in nearly three-quarters of test cases. Detection and patching scores were materially lower — indicating that AI is currently better at exploiting vulnerabilities than fixing them.
OpenAI announced $10 million in API credits for cyber defense work, with priority given to open-source software and critical infrastructure projects. The timing was not coincidental. The Moonwell incident demonstrated that AI tools can introduce vulnerabilities into production DeFi contracts, while EVMbench demonstrated that AI tools can also find and exploit them.
This creates a dual-use problem with no clear resolution. The same model architecture that generates vulnerable oracle code can also be deployed to scan for those exact vulnerabilities. The economic incentive structure favors offensive use: a single successful exploit yields millions, while defensive security remains a cost center.
Applying the economic value framework to Q1 2026's $138 million in exploit losses reveals where value destruction concentrates.
Direct loss bearers:
Indirect costs absorbed by the ecosystem:
Value redistribution (to attackers):
The pattern confirms that DeFi's security costs are almost entirely socialized to end users, while the economic value of successful exploits flows to attackers with near-zero friction. Insurance mechanisms cover less than 0.5% of TVL at risk, and audit expenditures provide probabilistic — not deterministic — protection.
Q1 2026 data shows DeFi's security problem is not improving. Losses increased year-over-year despite lower total crypto market activity and reduced TVL. The attack surface is expanding beyond smart contract logic into key management, governance processes, and now AI-assisted code generation.
The Moonwell incident is particularly instructive. The vulnerability was not in the protocol's core contracts but in a governance-approved configuration change, co-authored by an AI tool, that bypassed existing audit coverage. As AI coding agents become standard development tooling — the Tenzai study tested tools already widely adopted across the industry — the volume of context-dependent security errors entering production code is likely to increase.
OpenAI's EVMbench and $10 million defensive commitment represent early-stage responses to this dynamic, but the economic incentives remain misaligned. Offensive exploitation is immediately profitable. Defensive security spending is a recurring cost with uncertain returns. Until insurance penetration, formal verification tooling, or regulatory requirements alter that equation, DeFi's $105 billion TVL will continue to subsidize a $100M+ quarterly transfer of value from depositors to attackers.