3.66 million transactions. 63% of them malicious. Over $2.36 million drained directly, with another $10.14 million exposed. These aren't from a buggy DeFi farm—they're the first three months of EIP-7702 on mainnet, the upgrade meant to bring smart contract powers to every EOA.
Audit complete. The soul remains. But the soul of EIP-7702—the promise of seamless, native account abstraction—is now bleeding through a thousand small cuts. The upgrade that was supposed to eliminate the friction of migrating to smart wallets has instead become a mass exploitation pipeline.
Let me be clear: I'm not against account abstraction. I've built governance frameworks that rely on it. But the numbers from the USENIX 2026 study—which I've been digging through for the past week—tell a story that the Ethereum Foundation's blog posts conveniently avoid. This isn't just a security issue; it's a fundamental challenge to the trust model of the world computer.
Context: The Architecture of Delegation
EIP-7702, included in the Pectra upgrade (activated May 7, 2025), allows an EOA to delegate its code to a smart contract without changing its address. Think of it as giving your old house a new foundation without moving the mailbox. The address stays the same, but now it can execute arbitrary logic, batch transactions, and sponsor gas. Beautiful in theory.
Digging deep for the truth in the chain.
The problem is that this delegation is a binary switch: once you sign a delegation message, your entire account becomes a puppet of the delegated contract. The private key still controls the address, but the code controls the semantics. In traditional EOA security, the private key is the ultimate authority. In EIP-7702, the private key is just a gatekeeper that can be tricked into signing a delegation that hands over the keys to the kingdom.
Core: The Numbers Don't Lie
Over 3.66 million EIP-7702 delegation transactions occurred in the first three months, according to the study that analyzed 228 billion historical transactions. Of those, 63% were associated with malicious contracts. The attackers didn't bother with sophisticated exploits—they used classic phishing, social engineering, and fake UIs to trick users into signing delegation messages.
But here's the kicker: 242 distinct malicious contracts were identified, and 500 more were deployed using CREATE2 (meaning they can be instantiated at will). The researchers found that the delegation code itself is often benign-looking—until it's triggered by a specific condition, like a transaction to a high-value address. Then it re-delegates to a different contract, effectively locking the user out of their own account.
Based on my audit experience—back when I built EthGuard Lite in 2017 for ERC-20 reentrancy—I know that the most dangerous vulnerabilities are the ones that break existing assumptions. EIP-7702 breaks the assumption that msg.sender == tx.origin is a reliable anti-phishing check. In fact, it breaks tx.origin entirely. Any DeFi protocol using that pattern (and many still do) is now vulnerable to a new class of attacks where the attacker's contract acts as a proxy.
The total confirmed losses: $2.36 million stolen. But the exposed value—funds in accounts that interacted with malicious delegations—is $10.14 million. And that's just the tip of the iceberg. The study didn't even account for the 500 un-deployed CREATE2 contracts that could be triggered at any moment.
Contrarian: The Upgrade Isn't the Enemy—It's the Speed of Trust
Here's the uncomfortable truth: EIP-7702 is not a bug; it's a feature being exploited before its time. The account abstraction community has been waiting for a native solution for years. The alternative—migrating every user to a new smart contract wallet—is logistically impossible for the Ethereum base layer. EIP-7702 offers a migration path that preserves address continuity, which is critical for DeFi positions, NFTs, and reputation systems.
The contrarian angle is that the attack vector is not a failure of the protocol but a failure of the ecosystem to adapt fast enough. The wallets (MetaMask, Rabby, etc.) implemented the delegation UI without proper warnings. The security tools (block explorers, simulation services) didn't update their heuristics to detect malicious delegations. The users were left to sign arbitrary messages without understanding the consequences.
Archaeologists of the abstract. We're digging through the rubble of the first wave of adoption, and what we find is a pattern: every major upgrade to Ethereum's account model (from EIP-1559 to EIP-3074) has faced a similar trust crisis. The difference is that EIP-7702 operates at the level of the account itself—the most fundamental unit of ownership. The stakes are higher.
Yes, the protocol is sound. The math works. But the human layer didn't catch up. And until it does, every signing ceremony is a potential loss of sovereignty.
Takeaway: The Security Oasis Must Be Built Before the Next Caravan
The USENIX paper is a wake-up call, but it's not a death knell. The solution is not to roll back EIP-7702 (that ship has sailed) but to build a new security infrastructure around it. I'm talking about:
- Delegation whitelists at the wallet level (only allow delegation to audited contracts)
- Real-time taint tracking for delegated accounts (if the code changes, the wallet should flash red)
- On-chain reputation systems that mark known malicious delegation contracts
Audit complete. The soul remains. The soul of account abstraction is still alive—it's just bruised. The question is whether the Ethereum ecosystem can learn from 3.66 million transactions that 63% were malicious. If we can't, we're not just losing millions; we're losing the trust that makes decentralization possible.
The next 12 months will determine whether EIP-7702 becomes a footnote in crypto's history of failed upgrades or the stepping stone to a truly programmable ownership model. I'm betting on the latter, but only if we stop pretending that technical upgrades are enough. We need an emotional upgrade—a collective understanding that every signature is a commitment, and every delegation is a leap of faith.