Wallets
Address poisoning: the lookalike in your history
Address poisoning needs no access to your keys, your device or your approvals. It works on one habit, copying a recipient’s address from your own transaction history, and on how few characters of an address anyone reads. The scammer makes an address that matches the ends of one you use, gets it into your history, and waits for you to pick it.
The habit it relies on
An address on Ethereum, and on chains that share its format, is 40 hexadecimal characters. A study presented at USENIX Security 2025 by a team mostly from Carnegie Mellon University, which found 13 times more attack attempts than earlier work had reported, starts from that: because the address is “a hardly memorable 40-digit hexadecimal string”, users “often select addresses from their recent transaction history” (paper).
Interfaces shorten addresses as well. When the researchers checked in January 2025, MetaMask and Phantom showed the first four and last four characters and Trust Wallet the first two and last four, while Etherscan kept the same number of characters at each end and dropped the middle. An attacker does not need your counterparty’s address. They need one that agrees with it where you look.
Making a lookalike
An address is derived from a public key, so the attacker generates key pairs until one produces an address with the right ends. Each hexadecimal character takes one of 16 values, so matching four characters at each end takes about 4.3 billion attempts on average (16 to the eighth power), and each further character multiplies the work by 16. That is within reach of graphics hardware, and the paper identifies “a large-scale attacker group that appears to use GPUs”. For their own controlled test the researchers generated a lookalike matching the first six and last four characters of the target.
Ethereum’s checksum does not help. EIP-55 encodes a check in the capitalisation of an address and estimates that a mistyped address passes it 0.0247% of the time. That protects against typing errors. A lookalike is not a typo: it is a different, valid address with a valid checksum of its own.
Getting it into your history
The paper describes three kinds of poisoning transfer, which attackers often combine. In a tiny transfer, the lookalike sends you a small amount of the token you recently sent your real counterparty, so it appears in your history beside them.
A zero-value transfer goes further. The attacker records a transfer from your address to the lookalike, of the token you just sent, with a value of zero. ERC-20 requires that “transfers of 0 values MUST be treated as normal transfers and fire the Transfer event”, and, in the paper’s words, “popular token implementations allow the function callers to specify an arbitrary sender and an arbitrary receiver if the transferred value is zero”. The researchers rate these “potentially even more dangerous than tiny transfers, as the victim appears to have previously sent money to the lookalike address”.
The third kind uses a counterfeit token. The attacker creates a token of its own with a similar-looking name and “transfers” it from you to the lookalike, for the amount you sent your counterparty. The attacker wrote the token contract, so it can move that token on anyone’s behalf.
The paper records attackers putting more than 100 poisoning transfers into a
single transaction, and finds that in 79% of successful attacks on Ethereum more
than one lookalike was aimed at the same victim. Blocking zero-value
transferFrom in existing tokens has been proposed and dropped. A 2023 request
on OpenZeppelin’s contracts repository to disallow
it was
closed, and the researchers summarise the outcome: no change, because removing
zero-value transfers “would break existing code”.
A $68 million example
One large case was reconstructed on-chain by Chainalysis, a blockchain analytics firm, in October 2024, and the account here is theirs. At 09:14 UTC on 3 May 2024 the victim sent a transfer to an address beginning 0xd9A1b, which Chainalysis reads as “likely a test payment”. Minutes later the victim sent about $68 million in wrapped bitcoin to 0xd9A1c, a scammer’s address that “looks to be the same ‘0xd9A1,’ if you only reference the first six characters”. The USENIX paper gives the amount as 1,155 WBTC. By 14:44 UTC the scammer had moved the funds on. Over the following days the victim sent messages embedded in small ether transactions, negotiating, and on 9 May the scammer returned the original $68 million in ether, keeping about $3 million that had accrued as the price rose.
Two details matter more than the amount. The victim had done the cautious thing first: the test proved that one address worked, and the next transfer went to a different one. And the wider campaign was not aimed at beginners. Chainalysis linked 82,031 potential seeded addresses to it, “slightly under 1% of all new Ethereum addresses created during the period of the scam campaign”, and found that its victims’ wallets were older, busier and larger than typical ones.
How common it is
The USENIX study measured two years of activity on Ethereum and BSC and found “270M on-chain attacks targeting 17M victims”, with 6,633 incidents causing “at least 83.8M USD in losses”. It tracked only a few stablecoins, so the authors call the loss figure a conservative lower bound, and the WBTC case falls outside it. In January 2026 Carnegie Mellon’s CyLab reported that the team’s monitoring tool had detected “a successful $50 million attack on December 19, 2025”.
Very little of this reaches the hacks tables. DefiLlama’s record, read in full in how crypto gets hacked, lists incidents at projects and venues, not individual losses, and as of 22 November 2026 it carries one incident with the technique Address Poisoning, filed under Social Engineering.
What stops it
The defences come down to never choosing an address from history, and checking what you send against where the address came from.
- An address book filled from the source. Save a counterparty’s address once, from where it originates (an exchange’s deposit screen, the person’s own message, confirmed through a second channel) and send from the saved entry. The researchers list an allow-list of trusted addresses among user-level defences, and CyLab’s summary of their advice includes “whitelisting trusted addresses when possible”.
- The whole address, against the original. The ends are what the attacker matched. Compare the full string, middle included, with the source and not with your history.
- A test transaction, then the same address. A small first payment proves the address you tested. Chainalysis found that 756 of the campaign’s addresses had received between $1 and $100, “likely test payments”, and that because those did not reach the intended address they prevented further losses. The main payment has to go to the address that was tested, copied again from the same source. A tiny poisoning transfer mirrors a payment you have just made, so a test can itself prompt a fresh lookalike to appear in your history.
- Unexpected incoming transfers as a warning. A zero-value or tiny transfer from an address resembling a counterparty is the attack in progress. Some wallets hide them: the researchers found that Trust Wallet and Phantom hid zero-value and counterfeit-token transfers in January 2025.
A hardware wallet’s screen confirms that the address being signed is the one you entered, and no more; how hardware wallets work quotes Trezor’s own statement of the limit, that the display “can’t tell you whether an address someone gave you is the right one”. Clipboard malware, which swaps an address between copy and paste, is a different attack with the same result, and how crypto wallets actually fail covers it. The routine in your first crypto wallet, a small send and then a send back, is the same habit as the third point.
After a mistaken transfer
A transfer to a lookalike is a valid transaction to an address someone else controls, and nothing reverses it. In one case the researchers observed, speed at the issuer helped: for the largest payoff transfer they found, “the Tether team (USDT) blocked the attacker’s address within an hour”, which stopped the attacker moving the funds on. That is possible only for assets whose issuer can freeze balances. The WBTC victim’s negotiation worked once and is not a method.
Anyone who then offers to recover the funds for a fee belongs to a different fraud, set out in crypto recovery scams. Theft by signature, where the attacker takes tokens rather than waiting to be paid, is the subject of wallet drainers explained.