TLDR
A peer-reviewed study for USENIX Security ’26 found attacker-linked contracts tied to 63% of Ethereum’s EIP-7702 authorization transactions.
Researchers tracked 3.66 million authorization transactions across seven chains through July 2025.
Detected losses totaled $2.36 million, with another $10.14 million in assets exposed through outdated contract defenses.
Attackers were found rebinding accounts to normal-looking code after attacks, making it harder to spot ongoing risk.
Researchers and Ethereum developers are now pushing wallets to vet and clearly display any code an account delegates to.
A new study has found that attackers were behind most of the early activity tied to a new Ethereum wallet feature. The research was peer-reviewed and presented for USENIX Security ’26.
The feature in question is EIP-7702, which lets a regular Ethereum wallet temporarily act like a smart contract. It went live as part of the Pectra upgrade on May 7, 2025.
Researchers studied more than 22.8 billion transactions across seven blockchains, including Ethereum, Binance Smart Chain, Polygon, Optimism, Arbitrum, Base, and Gnosis. Within that data, they found 3,664,166 EIP-7702 authorization transactions through July 15, 2025.
Of those, 2,322,548 transactions, or 63%, were linked to contracts the researchers identified as malicious. The team used transaction filters, code analysis, and manual review to confirm 924 malicious contracts overall.
How the delegation feature works
EIP-7702 allows a wallet address to point to separate contract code without changing the address itself. The original owner keeps their private key, but the linked code can act with the full authority of that account.
This setup lets wallets add features like batching multiple actions into one transaction or letting someone else pay the gas fee. It also means the linked code becomes part of the wallet’s security.
If that code is faulty or written by an attacker, it can approve transfers, move funds, or interact with apps as if it were the account owner. Researchers say attackers have prepared these authorizations ahead of time and gotten victims to sign them, sometimes through a wallet prompt that does not clearly show which code is being approved.
What the losses looked like
The study measured $2,362,848.76 in confirmed losses across three attack types. A separate part of the research looked at older contracts that assumed a wallet address could never behave like a contract.
That assumption broke once EIP-7702 launched. Researchers found 967 active Ethereum contracts still relying on it as a security check, exposing about $10.14 million in assets to potential risk.
Researchers also found attackers switching an account back to normal-looking code after an attack. This makes it hard for anyone checking a wallet’s current state to spot that it was compromised earlier.
They additionally found 500 delegation targets pointing to addresses with no code yet deployed. Code could be added to those addresses later, changing what the wallet actually does while its recorded target stays the same.
The study’s authors say their method may not catch every malicious contract, especially newer ones or those using different attack methods. The 924 confirmed contracts represent what they were able to verify, not the full scope of abuse.
In response, Ethereum.org has released guidance recommending wallets whitelist delegation contracts, clearly show users which code they are approving, and rely only on audited smart account implementations. A related proposal calls for wallets to stick to a short list of publicly reviewed account systems rather than letting any application request custom delegation code.
