Research
EIP-7702 Phishing Attack
Ethereum's dominant account type, the externally owned account (EOA), has fixed validation logic, so past phishing research has focused on tricking users into approving individual harmful…

- arXiv
- 2512.12174
- Published
- 2025-12-13
- Authors
- Minfeng Qi, Qin Wang, Ruiqiang Li, Tianqing Zhu, Shiping Chen
AI summary
Overview
Research area: Blockchain security — specifically Ethereum account abstraction, the EIP-7702 delegation mechanism, and transaction-level phishing.
Technical level: Intermediate to Advanced. The paper assumes familiarity with Ethereum account models (EOAs vs. contract accounts), typed transactions, RLP encoding, ECDSA signatures, and ERC-4337 UserOp flows.
Scope (one sentence): The paper shows that a single signed EIP-7702 authorization tuple can hand an attacker persistent execution control over a victim's externally owned account, validates three ways that control can be triggered, and measures how the mechanism is already being used (and abused) across major EVM chains.
What This Paper Is About
Ethereum's dominant account type, the externally owned account (EOA), has fixed validation logic, so past phishing research has focused on tricking users into approving individual harmful transactions — malicious calldata, infinite token approvals, fake NFT marketplaces, and so on. EIP-7702 changes this by letting an EOA sign a small authorization tuple that permanently points its code at a delegate contract, meaning every future call to that account runs attacker-supplied logic. The paper's goal is to show this is a qualitatively new phishing primitive, to demonstrate end-to-end that it produces full account takeover, and to measure how far real-world adoption has already drifted toward criminal infrastructure.
Key Contributions
-
First data-driven measurement of EIP-7702 delegation activity. The authors analyze over 150k authorization and execution events across 26k addresses and hundreds of delegator contracts on major EVM chains, characterizing the protocol's real-world footprint.
-
A demonstrated full execution hijack from one signature. They show that a single malicious authorization tuple converts a victim EOA into a persistent proxy resolving to attacker-controlled bytecode, and validate three independent trigger pathways — user-driven, attacker-driven, and protocol-triggered (ambient) — through end-to-end experiments in a patched local environment.
-
Two extended attack surfaces via composability. They show that ERC-4337's
EntryPointpipeline allows remote and repeated activation of the delegated code without further victim involvement, and that chain-agnostic authorization (chainId = 0) permits replay-like compromise across independent networks. -
Protocol-level and ecosystem-level defenses. The paper concludes with proposed mitigations to constrain the delegation-based phishing surface.
Main Findings
-
A single tuple replaces a stream of phishing approvals. In prior phishing families (payload-based, approval-based, permit-based, proxy-based, NFT
setApprovalForAll, DeFi router impersonation, MEV bait), the victim must authorize a harmful transaction or typed signature. Under EIP-7702, the victim signs a compact tuple(cid, addr, ν, y, r, s)that reveals nothing about the behavior it installs, yet redirects all future execution. -
The delegation write is not reverted on execution failure. The paper stresses that
code(auth) ← 0xef0100 || addris applied before the transaction's execution phase (after the sender nonce increment), and the write persists even if the later execution reverts. Attackers can therefore embed the tuple in an outer transaction that is a dummy call, a benign interaction, or one deliberately structured to fail. -
Wallets do not surface the authorization list. Because clients render only outer transaction fields, the tuple stays hidden inside the payload, letting the attacker present an operation consistent with familiar UX patterns such as a harmless approval or verification message.
-
Three trigger pathways were identified and tested. User-driven activation (the victim's own routine transaction, e.g., sending ETH to their own address to force-send gas or reset a stuck nonce), attacker-driven activation (the attacker calls the victim's address directly, even while the victim is offline), and ambient execution (token callbacks, settlement routines, or automated service calls that happen to target the victim).
-
Real-world delegation is highly centralized and crime-linked. The measurement shows burst-driven growth, extreme centralization dominated by a small number of contract families linked to criminal activity, and repeated reuse of delegator contracts across incidents. Loss data reveals substantial theft of ETH, ERC-20 tokens, and NFTs. (The provided excerpt does not report specific loss totals or per-chain breakdowns.)
-
Adoption is large enough that the risk is not theoretical. The authors report 2–3 million delegations per day observed on major EVM chains, and argue many real-world delegator endpoints function as malicious infrastructure already operating at scale.
-
Design-level enablers, not implementation bugs. The attack exploits two deliberate EIP-7702 properties — persistence of delegation across transactions until explicitly revoked, and
chainId = 0cross-chain applicability — combined with the opacity of the delegation pointer0xef0100 || address.
Methodology in Plain English
The authors combine formal modeling, a controlled testbed, and on-chain measurement.
First, they formalize EIP-7702: a set-code transaction (type 0x04) carrying an authorization list, where each tuple is validated against chain ID, the authorizing account's current nonce, and ECDSA canonicality, then ecrecover identifies the signer. They define a state transition that writes a delegation indicator into the account's code slot, and an attack model with a victim EOA, an attacker, and auxiliary infrastructure (wallets, relayers, bundlers). They then define the condition for a "successful drain" as a measurable drop in the victim's balance paired with value flowing to the attacker above a threshold Δ.
Second, they build the attack. The malicious delegate contract puts its draining logic in the fallback function, because under delegation the victim's address becomes msg.sender and every entry point is implicitly authorized. The fallback enumerates balances with balanceOf, issues transfer/transferFrom for ERC-20s, moves NFTs via transferFrom, optionally swaps assets to ETH through attacker-controlled routers, and forwards ETH with low-level calls propagating all available gas. Attackers are described as pre-deploying reusable templates for different extraction strategies and reusing the same target address across many victims.
Third, they implement and run it locally. Because Hardhat does not natively implement transaction type 0x04, they add a preprocessing layer (patchEVM.js, patchHardhatNode.js) that intercepts type-4 transactions, extracts and validates the authorization list, computes the authority address, and writes 0xef0100 || target before execution. Tooling includes generateTuple.js, sendAuthTx.js, trigger.js, and runAndCollect.js, with debug_traceTransaction used to instrument each scenario.
Finally, they measure the real world: over 150k authorization and execution events, 26k addresses, and hundreds of delegator contracts, examining who controls delegator contracts, how widely delegation is deployed, and whether observed endpoints already show malicious patterns.
Why This Matters
Impact on research. The paper argues there is a "missing link" in the phishing literature: all prior work assumes the victim authorizes a harmful transaction, whereas EIP-7702 lets a victim authorize a lasting change to account semantics. This reframes phishing from a transaction-level problem into an account-configuration problem, and it is claimed as the first security analysis of EIP-7702 as a phishing vector — a necessary complement to the protocol's rapid adoption.
Real-world applications.
- Wallet and signing-UI design: wallets must decode and display authorization lists and delegate contract behavior, not just outer transaction fields, before a user signs.
- Block explorer and node tooling: the finding that delegation writes are
revert-safe and applied during preprocessing means monitoring tools must flag type-0x04transactions regardless of whether the outer call succeeds. - Incident response and forensics: the observed reuse of the same delegator addresses across victims and incidents gives investigators a clustering signal.
- Chain and protocol governance: the centralization finding informs decisions about whether
chainId = 0authorizations and unrevoked delegations need policy-level constraints.
Industry relevance. With 2–3 million delegations per day on major EVM chains, EIP-7702 is not a fringe feature. Exchanges, custodians, and dApp front-ends that display signing requests need to treat authorization tuples as a distinct, high-risk signature type. The paper's evidence that delegator contract families are already linked to criminal activity suggests defenders are behind the curve.
Future Directions
-
Design and evaluate the proposed defenses. The paper closes by proposing protocol-level and ecosystem-level mitigations, but the excerpt does not specify their exact form or measure their effectiveness — leaving evaluation work open.
-
Extend the measurement. The reported analysis covers over 150k events, 26k addresses, and hundreds of delegator contracts. Tracking delegation growth, revocation rates, and delegation lifetimes over time, and attributing losses per incident, would sharpen the risk picture.
-
Quantify the ERC-4337 and cross-chain amplification. The
EntryPoint-based remote activation and thechainId = 0replay-like compromise are argued conceptually and framed as extended surfaces; systematic measurement of how often these paths are actually exercised in the wild remains open. -
Harden client and wallet validation. Since the attack depends on the gap between what users see and what the protocol interprets, concrete proposals for rendering authorization tuples, warning on unverified delegate targets, and flagging reusable template addresses are natural next steps.
Target Audience
Blockchain security researchers and graduate students working on transaction-level attacks and account abstraction; Ethereum protocol and client developers who need to reason about EIP-7702's threat model; wallet, explorer, and custody-product engineers responsible for signing UX and transaction simulation; and threat-intelligence or incident-response teams tracking approval-based and delegation-based theft. Readers without prior exposure to Ethereum's account model and typed transactions will need background reading to follow the formal sections.
Authors’ abstract
EIP-7702 introduces a delegation-based authorization mechanism that allows an externally owned account (EOA) to authenticate a single authorization tuple, after which all subsequent calls are routed to arbitrary delegate code. We show that this design enables a qualitatively new class of phishing attacks: instead of deceiving users into signing individual transactions, an attacker can induce a victim to sign a single authorization tuple that grants unconditional and…
The full abstract on arXiv