# Blockchain Transaction Simulation Phishing

Crypto.world

> Crypto wallets such as MetaMask, Coinbase, and Trust Wallet now "simulate" a transaction before a user signs it, showing the expected balance change so users can spot fund-draining scams.

**arXiv**: 2607.28747

**Published**: 2026-07-30

**Authors**: Xiaocan Wang, Shixuan Guan, Tong Yang, Xiapu Luo, Yue Duan, Kai Li

## AI summary

## Overview

**Research area:** Blockchain security (cs.CR) — crypto-wallet phishing, smart contract program analysis, and measurement of attacks on EVM-compatible chains.

**Technical level:** Intermediate. The attack story and measurements are accessible to a general security audience; the detection design (symbolic execution, three-address-code IR, constraint solving) requires some program-analysis background.

**Scope:** The paper defines, characterizes, and measures at scale a new phishing technique called *transaction simulation phishing*, in which smart contracts behave differently under wallet transaction simulation than during on-chain execution, and it introduces a bytecode-level detector named SimGuard.

## What This Paper Is About

Crypto wallets such as MetaMask, Coinbase, and Trust Wallet now "simulate" a transaction before a user signs it, showing the expected balance change so users can spot fund-draining scams. The paper shows that this defense can be turned against users: carefully crafted smart contracts make the simulation show a benign or even profitable outcome, while the real on-chain execution sends the user's funds to an attacker-controlled address. The goal is to understand how these contracts work, build a tool to find them across blockchains, and quantify how many contracts, victims, and dollars are involved.

## Key Contributions

1. **First comprehensive study of transaction simulation phishing.** The authors characterize a new attack that exploits the simulation feature itself, showing it bypasses a defense designed to stop traditional phishing.
2. **A taxonomy of six phishing contract classes.** Based on a systematic analysis of which blockchain property variables can be used as control variables, they define *storage-control*, *external-control*, *Gas-control*, *Gasprice-control*, *blocknumber-control*, and *timestamp-control* contracts.
3. **SimGuard, a bytecode-level detection system.** SimGuard combines symbolic execution into a three-address-code representation, phishing-specific recognition of suspicious transfer patterns, data-flow tracing of the control variable, and runtime verification; it reports zero false positives on the authors' ground-truth dataset and 97.72% recall.
4. **Large-scale measurement across four blockchains.** Applying SimGuard to Ethereum, Binance Smart Chain, Avalanche, and Polygon yields more than 4,000 phishing contracts, more than 5,700 victim addresses, and approximately $3.48 million USD in losses, plus timeline and attacker-clustering analysis. The authors state that SimGuard and the collected dataset have been open-sourced.

## Main Findings

- **Detection volume:** 4,224 phishing contracts were detected across the four chains, broken down in the paper's Table 2 as 480 on Ethereum, 293 on BSC, 3,136 on Avax, and 315 on Polygon. (The abstract states 3,316 for Avax; the per-chain counts in Table 2 sum to 4,224.)
- **Gas-control dominates:** 4,192 detected contracts use Gas-control, which the authors describe as 99.2% of all detections. The remainder is 22 storage-control and 10 timestamp-control contracts. External-control, Gasprice-control, and Blocknumber-control contracts were not observed in the wild, for reasons the authors attribute to implementation complexity, threshold instability, or extra operational overhead.
- **Avalanche is the most affected chain:** over 3,100 contracts there (74.2% of detections), while Ethereum contributes the largest share of losses.
- **Victims and losses:** more than 5,700 victim addresses and approximately $3.48 million USD in losses, with 91.5% of the losses occurring on Ethereum ($3.19M, from 1,104.70 ETH). Table 3 reports 6,223 victim transactions in total and 146 "lucky" transactions where users received back more than they deposited.
- **Storage-control is the most profitable class:** over $2.97 million USD, or 85.34% of total losses, which the authors attribute to its deterministic control over the execution outcome.
- **Time trend:** the technique first appeared in August 2024; activity intensified between March and May 2025; Avax saw more than 550 contracts deployed in a single day on April 22, 2025; on May 3, 2025 all four chains together exceeded 100 deployments in one day; the most severe losses were concentrated early in the campaign, including a single day in January 2025 with losses exceeding $1 million USD.
- **Extreme attacker returns:** return on investment ranged from 4,463% to 83,495% across the four blockchains.
- **Coordinated attacker groups:** clustering shows a small number of groups dominate; the largest Ethereum cluster alone accounts for approximately $2.88 million USD, or about 83% of total profits.
- **Explorer labeling lags:** of the 4,224 detected contracts, 899 carry a "phishing" label from the relevant explorer and 3,325 do not. On Avax only 12 of the detected contracts were flagged, versus 3,124 unflagged.
- **Detector accuracy:** on a ground-truth set of 44 phishing contracts (14 real storage-control contracts found by snowball sampling plus 30 Gemini-synthesized contracts covering all six classes) and 162,975 benign contracts (139,451 from Forta plus 23,524 expanded by the authors), SimGuard achieved 33 true positives, 1 false negative, 0 false positives, precision of 100%, and recall of 97.72%. Average analysis time was 4.38 s per phishing contract and 6.50 s per benign contract.

## Methodology in Plain English

The authors start by enumerating which parts of a blockchain transaction a contract can observe and which of those an attacker can predictably influence. They reason that transaction fields the simulator may change (Gas limit, Gas price) and monotonic block values (block number, timestamp), plus mutable contract storage and external contract state, can all be used as a switch: one side of the switch returns the user's deposit plus a small "profit," the other side forwards the deposit to the attacker. They illustrate this with a contract whose `claim()` function pays the caller back unless the caller has been blacklisted, and they explain the attack as a time-of-check-to-time-of-use (TOCTOU) race: the attacker front-runs the victim's transaction to flip the contract's state or to make the block number, timestamp, Gas, or Gas price differ from what the simulator used.

To find such contracts without source code, SimGuard pulls deployed bytecode from historical transaction data using Ethereum-ETL, filters out well-known benign contracts, and deduplicates identical code. It then runs the existing `greed` symbolic executor to explore all branches and produce a three-address-code intermediate representation. From that representation it looks for a `JUMPI` whose two successor blocks both contain a `CALL`, where one `CALL` sends the caller more than the deposited value (`CALLVALUE`) and the other sends the deposit to an address loaded via `SLOAD` or returned by another `CALL`. It then traces what that branch condition depends on (storage, calldata, `tx.gas`, `tx.gasprice`, `block.number`, `block.timestamp`) to assign the contract to a taxonomy class. Finally, each suspicious contract is executed in a local Foundry environment with two groups of test transactions — one meant to trigger the profit path and one meant to trigger the loss path — with a constraint solver choosing parameters to reach the intended code locations. Only contracts that exhibit both behaviors are confirmed as phishing.

## Why This Matters

The paper shows that a widely deployed defensive feature can create its own attack surface, which shifts the research question from "how do we warn users about bad transactions" to "how do we make the warning itself trustworthy."

**Real-world applications:**

- Wallet vendors (MetaMask, Coinbase, Trust Wallet and others) can use the taxonomy to harden simulation — for example by re-simulating at signing time, pinning block parameters, or flagging contracts whose outcome depends on mutable state or caller-supplied Gas.
- Blockchain explorers and threat-intelligence teams can feed SimGuard-style bytecode detection into address labeling, which matters given that 3,325 of 4,224 detected contracts were still unflagged.
- Incident responders and law enforcement analysts can use the clustering and loss-attribution results to identify coordinated campaigns and the largest profit-taking clusters.
- Users and wallet UX designers get a concrete lesson: a "profitable" simulation preview is not proof of safety, especially when the transaction is initiated from a site promising free cryptocurrency.

**Industry relevance:** The paper reports that crypto losses in the first half of 2025 alone exceeded $2.17 billion USD, and it documents a live campaign running from August 2024 onward with attacker ROIs in the thousands of percent — strong economic incentive for defenders to act.

## Future Directions

- **Stronger simulation semantics.** Designing wallet-side simulation that is robust to state changes between simulation and execution, and that surfaces when a contract's outcome depends on storage, Gas, block number, or timestamp.
- **Closing the obfuscation gap.** The single missed detection was a Gasprice-control contract using opaque predicates, redundant loops, and indirect branch routing; handling such obfuscation is an open program-analysis problem.
- **Tracking attacker adaptation.** Three taxonomy classes were absent in the wild — understanding whether attackers migrate to them as defenses improve is an open measurement question.
- **Extending beyond the four chains and beyond detection.** The paper covers EVM-compatible chains up to August 2025; broader coverage, cross-chain attacker attribution, and countermeasures such as automated takedown or explorer labeling pipelines are natural next steps.

## Target Audience

Blockchain security researchers and graduate students studying phishing and smart contract analysis; wallet and exchange security engineers who build or rely on transaction simulation; blockchain threat-intelligence and explorer labeling teams; and program-analysis researchers interested in applying symbolic execution to adversarial, obfuscated real-world bytecode.

## Authors’ abstract

Cryptocurrency users have increasingly become targets of phishing and scam attacks. To mitigate these threats, leading crypto wallets (e.g., MetaMask) have introduced transaction simulation, which previews a transaction's balance changes before on-chain execution. While effective against traditional fund-draining attacks, we show that this defense can itself be exploited by a new phishing technique, which we term transaction simulation phishing. This attack uses carefully…

[The full abstract on arXiv](https://arxiv.org/abs/2607.28747)

[Read the original paper](https://arxiv.org/abs/2607.28747)

---

Canonical: https://crypto.world/paper/blockchain-transaction-simulation-phishing-2607.28747
