1 of 27

Insecurity Through Obscurity: Veiled Vulnerabilities in Closed-Source Contracts

Sen Yang

DeFi Security Summit

Nov 20 2025

​

With Kaihua Qin, Aviv Yaish and Fan Zhang

1

2 of 27

Decentralized Finance

2

DeFi Application

Blockchain

3 of 27

MEV and MEV bots

3

Arbitrage

Liquidation

Sandwich

Other DeFi attacks

Identifying opportunity

Extracting value

MEV searchers

MEV bots

4 of 27

MEV bots

4

Screenshots from etherscan and eigenphi

Wealthy and Mysterious

5 of 27

Attacks against MEV bots

5

From 2021 to 2024 (or …?)

There is nothing new under the sun.

But, is it complete?

6 of 27

Overview

  • New attack vector: MEV phishing attacks
  • Challenges in identifying security issues
  • Our solution: SKANF

6

7 of 27

MEV phishing attacks

Why: the access control mechanism.

Ensuring security may be costly in blockchain systems.

7

~450 gas

8 gas

Should I use tx.origin?

8 of 27

MEV phishing attacks

Idea of MEV phishing attacks: be part of the MEV supply chain;

Thus, it can bypass tx.origin check.

​

​

​

​

8

Overall statistics:

    • We identify 104 attacks from July 2021 to April 2025
    • $2.76M in losses; the single largest attack: $636K
    • It includes three variants.

MEV Extraction

malicious tokens / pools / addresses

9 of 27

Token-based MEV-phishing attacks

Once searchers perform arbitrages involving a malicious token, the backdoor within the token is triggered to attack MEV bot.

​

​

​

​

9

10 of 27

Pool-based MEV-phishing attacks

Deploying a malicious pool to attract searchers to extract MEV.

10

11 of 27

Refund-based MEV-phishing attacks

Token and pool are not malicious. But searchers need to refund a malicious address if they want to backrun this opportunity.

11

12 of 27

Why it is hard to identify security issues?

MEV bots keep suffering losses.

12

Question: Why do they not use tools to secure their contracts?

😭

🤔

Or, what are the challenges to prevent them from being useful?

13 of 27

Challenge 1: control-flow obfuscation

  • Developers may introduce control-flow obfuscation intentionally or as a side effect of gas optimization.
  • Existing approaches no longer work in such smart contracts.

13

most of code cannot be analyzed by security tools

14 of 27

Challenge 2: the complexity of the contract

Even without obfuscation, the contract itself can still be very complex.

Analyzing tools may time out.

14

Logic is very complex

15 of 27

Challenge 3: pattern match cannot help

Is a partially controllable call vulnerable or not?

�This is hard to answer because we do not know what parts we can control or whether we can reach this call.

15

16 of 27

Idea: de-obfuscation

16

Control flow obfuscation: the jump destination is based on a runtime value (indirect jump)

Observation: EVM specifies that JUMPDEST should mark every legal jump destination

​

Therefore, we can manually collect all legal jump destinations and rewrite the instructions to make the control flow explicit.

17 of 27

Idea: prioritize exploration

  • A contract includes many execution paths
  • If we do a blind path exploration, it will be very slow for a complex contract.

​

17

…

State explosion

vulnerability

18 of 27

Idea: prioritize exploration

  • Observation: For smart contracts, we know their real-world execution traces!

​

18

Why do we not prioritize exploration based on these real-world traces?

19 of 27

Idea: prioritize exploration

We care about the vulnerability within the most frequent execution path.

19

…

historical transactions

vulnerability

20 of 27

Idea: identify all controllable / risky params

Combine taint analysis to identify all parameters that are controllable

​

20

SSSS

calldata

WETH.call

SSSS

a9059cbb

0000010000

Find a pattern that is not controllable but “risky”

transfer

Can be exploited by an adversary

21 of 27

Putting it together: SKANF

Symbolic Execution

21

Historical transactions

Smart Contract

De-obfuscation if it is obfuscated

Taint Analysis

Exploit

Potential Vulnerability

config (token, recipient, etc)

​

22 of 27

Evaluation: de-obfuscation

22

  • SKANF can significantly recover the control flow and thereby enhancing the efficacy of the static analysis tools.

23 of 27

Evaluation: vulnerability detection

  • Evaluation dataset: 6,554 MEV bots

​

  • Compared to other SOTA symbolic execution tools.
    • ETHBMC finds 0
    • JACKAL finds 18
    • Mythril finds 89
    • SKANF finds 1,030 vulnerable smart contracts
    • For 394 of them, SKANF successfully generate 394 exploits

​

    • We manually evaluate 100 of the cases where SKANF fails to generate exploits and find that exploit generation requires more sophisticated strategy.

23

24 of 27

Evaluation: vulnerability detection

  • If the vulnerable smart contracts were exploited, what would the potential loss be?

24

More than $10M

25 of 27

Evaluation: runtime

25

  • With the same time budget, SKANF can find more vulnerabilities than other state-of-the-art tools.

26 of 27

Summary

  • MEV-phishing attacks: token-based / pool-based / refund-based
    • New attack vectors against real-world MEV bots
  • Challenges of analyzing such contracts
    • Obfuscation / complexity / simple pattern match does not help
  • SKANF
    • De-obfuscation / historical transaction driven symbolic execution / taint analysis
    • Evaluation shows its efficacy in obfuscation and vulnerability detection.

​

26

27 of 27

Thank you!

Questions

27

Email: sen.yang@yale.edu / Twitter (X): syang2ng

arXiv link

blog post

https://syang.xyz