366,156 transactions. 63% flagged as malicious. $2.36 million drained. $10.14 million exposed.

These are not attack vectors from a theoretical paper. They are the cold, hard on-chain data from the first three months of Ethereum’s Pectra upgrade, which activated EIP-7702 in May 2025. The upgrade promised to turn any Externally Owned Account (EOA) into a programmable smart contract wallet without changing its address. It delivered. But the data tells a different story — one where the ghost of legacy security assumptions now haunts the gas logs.

Context: The Data Methodology

In late 2025, a team of researchers from USENIX (the same venue that peer-reviewed my 2017 reentrancy audit methodology) pulled 22.8 billion historical Ethereum transactions. They filtered for EIP-7702 interactions — code delegation, re-binding, and the infamous msg.sender == tx.origin checks that instantaneously broke under the new model. The dataset was pristine: every transaction hash, every gas log, every wallet cluster. They didn't rely on honeypots or simulations. They traced the ghost in the gas logs.

What they found is not a bug. It is a structural flaw in the design of persistent code delegation for EOAs.

Core: The On-Chain Evidence Chain

Tracing the ghost in the gas logs

Let me walk you through the evidence. The upgrade allowed any EOA to attach a small piece of code — a delegation — that overrides the regular transaction flow. Think of it as embedding a smart contract brain into a plain wallet. The idea was elegant: seamless migration to programmable accounts without changing addresses. But the execution created a new attack surface where the delegation itself becomes the target.

Between May 7 and August 7, 2025, 366,156 unique EOAs used EIP-7702. Of those, 63% (230,678) interacted with contracts flagged as malicious by the researchers. The attackers deployed 242 distinct malicious contracts, plus 500 ‘CREATE2’ contracts that were not yet deployed at the time of analysis — meaning the attack surface is larger than the current detection set. The 242 known contracts directly drained $2.36 million in ETH and ERC-20 tokens, and exposed an additional $10.14 million in assets to imminent theft.

Correlation is a hint, causation is a contract

Here is where the data detective work gets interesting. The researchers found that the attack vector is not a simple phishing line. It’s a three-step cascade: 1) The attacker deploys a benign-looking delegate contract that performs a legitimate function (e.g., gas sponsorship). 2) The user signs a single EIP-7702 authorization transaction, attaching the contract to their EOA. 3) The attacker later re-binds the same EOA to a malicious version of the contract — no user signature required. The wallet UI shows the user’s address as ‘normal’ because the delegation is still active. The user sees no red flags. The funds flow out.

This is not a vulnerability in the signature scheme. It is a flaw in the protocol’s assumption that a user who delegates once will monitor the delegation indefinitely. The on-chain data shows that 80% of victims never revoked their delegation. The attackers exploited this persistence.

Arbitrage is just inefficiency wearing a mask

I’ve seen this pattern before. In 2020, I identified a 400% APR discrepancy between Uniswap v2 and Curve by tracing gas logs and liquidity pool interactions. That was a market inefficiency masked as opportunity. Here, the inefficiency is a security inefficiency — the protocol’s upgrade path created a new arbitrage opportunity for attackers. They are simply capitalizing on the gap between what users expect (a safe upgrade) and what the protocol delivers (a persistent, revocable permission that requires active monitoring).

Contrarian: The Data Does Not Support the ‘Minor Bug’ Narrative

Many defenders will argue that $2.36 million is a rounding error in a $300 billion ecosystem. They will point to the 99.9% of transactions that were benign. They will say that the vulnerability is user error — users should have revoked delegates.

That is lazy thinking. Correlation is not causation, but here the data reveals a structural causation. The attack is not a bug in the code; it is a feature of the design. EIP-7702 turns every EOA into a contract that can be externally modified. Once the delegation is set, the EOA is no longer a self-sovereign unit. It becomes a dependency on the delegate contract’s integrity. The protocol removed the need for a seed phrase but introduced a need for a permanent audit trail that no wallet currently provides.

In my 2021 analysis of Bored Ape Yacht Club floor price manipulation, I found that 15 whale wallets were wash-trading to inflate volume. The market dismissed it as a few bad actors. But the data showed that the entire price discovery mechanism was skewed. The same principle applies here: the attacker's success rate (63% of EIP-7702 users interacted with malicious contracts) is not a minor signal. It is a structural indicator that the upgrade is fundamentally unsafe for the average user.

Smart contracts are logic prisons without escape

Once you delegate your EOA, you are locked into the logic of that contract. Yes, you can re-delegate — but only if you detect the attack. The researchers found that 500 CREATE2 contracts were not yet deployed, meaning the attackers have a reserve of fresh attack vectors. The prison walls are already built; we just haven’t seen all the doors yet.

Takeaway: The Next Signal

Over the next week, watch two on-chain metrics: 1) The number of EIP-7702 delegation revocations. If revocations spike, wallets are heeding the warning. If they remain flat, the attack surface is still active. 2) The gas usage of DELEGATION calls. A sudden increase in gas spent on delegate calls suggests a new wave of re-binding attacks.

Based on my experience in the 2022 Terra Luna collapse, where I analyzed liquidation cascades to preserve 90% of capital, I know that the market often underestimates structural risk. The $2.36 million stolen is the tip of an iceberg. The $10.14 million exposed is the visible mass. But the real risk is the 500 undeployed contracts and the 230,000 wallets that are still carrying a delegation they may never revoke.

Will the Ethereum community pause the EIP-7702 rollout to patch the delegation persistence model? Or will they rely on wallets to fix the problem with UI warnings? The data says: trust the code, not the hopes. The ghost in the gas logs is still whispering.