The numbers are stark. Over 366 million transactions. 63% of them malicious. $2.36 million in confirmed losses. Another $10.14 million exposed to risk. The data shows that EIP-7702, Ethereum's flagship account abstraction upgrade, is not a smooth evolution—it's a security minefield.
Context: The delegation model
EIP-7702 went live with the Pectra upgrade on May 7, 2025. It allows externally owned accounts (EOAs) to delegate their signing authority to a smart contract, effectively giving static addresses programmable logic. The promise: no more seed phrase migrations, seamless multi-signature, and gas abstraction. The reality: the same flexibility that enables innovation also enables exploitation.
A USENIX 2026 research paper, based on analysis of 228 billion historical on-chain transactions, mapped the attack surface. The study identified 242 distinct malicious contracts exploiting the delegation mechanism. The core vulnerability: the delegation model breaks the long-standing security assumption that msg.sender == tx.origin is a reliable check. This assumption was baked into thousands of DeFi contracts, yield vaults, and bridge protocols. Now it's worthless.
Core: The anatomy of the attack
Attackers moved fast. Within the first three months of activation, malicious delegation transactions accounted for 63% of all EIP-7702 activity. The numbers are not noise—they are targeted, automated exploits.
Three attack vectors dominate:
Re-binding attacks. Users sign a delegation to a trusted contract, say a legitimate gas-optimizer. Attackers intercept the delegation and rebind it to a malicious contract using CREATE2, before the user's transaction is mined. The user's EOA now delegates to a contract that drains their assets. The code does not lie, only the audits do. The user's wallet shows the original contract address, but the actual execution path is hijacked.
Broken standard checks. The EIP-7702 specification does not mandate that the delegated code is immutable or audited. As a result, 500 contracts were deployed via CREATE2 with no on-chain deployment transaction—they can be materialized at any time. This means the attack surface is still expanding. The research shows that over 80% of the malicious contracts were deployed using this method, making them invisible to static analysis tools.
Circular delegation. Attackers set up chains of delegations that loop back to the original EOA, creating recursive calls that bypass gas limits and reentrancy guards. The study found at least 15 such chains, each associated with drainer contracts that siphoned ETH and ERC-20 tokens.
Compounding the problem: the old guardrails are gone. Smart contracts that relied on tx.origin to prevent phishing—like many lending protocols that only allow the transaction originator to withdraw—are now wide open. The delegated code can spoof the tx.origin value, making these checks useless. The research estimates that $10.14 million in assets are currently exposed to this specific vulnerability across the top 10 DeFi protocols.
Contrarian: The 'safe by default' fallacy
A common narrative around EIP-7702 is that it's secure because it's a soft upgrade—the user's private key remains the ultimate control. The code is just a delegation, not a transfer of ownership. The data refutes this. The attack is not on the private key; it's on the signing context. The user signs a delegation, but the delegation is malleable. The attacker doesn't need the key; they just need to get the user to sign a blind transaction.
This is not a theoretical risk. The study's on-chain analysis shows that the median time between a delegation transaction being mined and the first malicious interaction is 1.2 seconds. That's faster than any human can react. The assumption that users can 'just revoke' is operationally false. The code executes logic, not intentions.
Furthermore, the research highlights that the percentage of malicious transactions has been increasing month-over-month, from 48% in May 2025 to 71% in July 2025. Attackers are learning, adapting, and scaling. The smart money is not in EIP-7702—it's in shorting the security of any protocol that has not revisited its tx.origin checks.
Takeaway: Actionable levels
This is not a theoretical paper. The losses are real. The data is on-chain. The countermeasure is clear: every project that uses msg.sender == tx.origin must audit and replace it with a whitelist-based delegation verification. Wallets must implement mandatory delegation co-signing—a second signature approving the exact target contract address. The Ethereum Foundation should consider a future EIP that mandates a canonical delegation registry, akin to a DNS for smart contracts.
Until then, treat every EIP-7702 transaction as a potential exploit. The code does not lie, only the audits do. The code says 63% are malicious. The question is: how many more will you sign before the industry catches up?