Address substitution attacks, address-book poisoning, and clipboard hijackers all exploit the same weakness: users compare only the first and last few characters of a cryptocurrency address instead of every symbol. A new open-source library called Humanized Hash (hh) converts any address or hash into a 4×4 grid of colored geometric shapes that is far harder to forge convincingly.
Real-World Incident: 1,155 WBTC Stolen in May 2024
On 3 May 2024 an attacker sent 1,155 WBTC (approximately 68 million dollars at the time) after poisoning the victim’s transaction history. The attacker first received a 0.05 ETH test payment to the correct address, then sent a small amount from a look-alike address beginning with 0xd9A1 and ending with 853a91. When the victim later copied the recipient from history, two nearly identical entries appeared and the wrong one was selected. The attacker later returned roughly 22,960 ETH after on-chain negotiations, but most victims never recover funds.
Clipboard Hijackers and In-App Malware
Clipboard malware, also called clippers, monitors the system clipboard and replaces any detected crypto address with an attacker-controlled substitute. Early samples replaced addresses entirely, but modern variants such as Laplas Clipper generate look-alikes that match the first and last characters. In September 2025 malicious versions of the chalk and debug npm packages were published after a maintainer account was compromised; the payload rewrote addresses inside browser fetch requests using Levenshtein distance. ReversingLabs later found a package that patched Atomic and Exodus wallets directly on disk. Dr.Web reported modified Android phones shipping with fake WhatsApp that swapped addresses between sender and receiver screens.
Address Poisoning on the Blockchain
Attackers also send tiny “dust” transfers or zero-value ERC-20 events from look-alike addresses. Carnegie Mellon University researchers documented 270 million poisoning transactions over two years on Ethereum and BSC, resulting in at least 83.8 million dollars in losses. After the Fusaka upgrade lowered gas fees, Blockaid recorded 3.4 million poisoning attempts in January 2026 alone. One campaign analyzed by Chainalysis seeded 82,031 look-alike addresses and remained profitable even after returning the largest theft.
Why “Check Every Character” Fails
Wallet vendors including Trezor and Ledger advise users to verify full addresses, yet eye-tracking studies show 38 percent of participants miss mismatches when only the first and last ten characters differ. MetaMask itself acknowledged that truncation creates the exact blind spot attackers exploit. GPU-accelerated tools such as profanity2 can generate four-character prefix and suffix matches in seconds on a single RTX 4090.
Limitations of Existing Identicons
Jazzicon in MetaMask is seeded from the first eight hex characters, so any address sharing those bits receives an identical icon. Blockies can also be matched within 45 minutes. Research by Tan et al. demonstrated that users comparing cute avatar-style images miss attacks 54 percent of the time because they judge overall appearance rather than individual features.
Humanized Hash Design and Security Analysis
The hh library maps any 256-bit input to a 4×4 grid where each cell is either empty or contains one of four shapes in one of four colors. Every cell depends on every input bit, and the algorithm applies 16,384 PBKDF2 iterations. A perfect visual forgery requires approximately 2^68 attempts; allowing two differing cells still demands roughly 2^52 attempts after stretching. Combined with matching address prefixes and suffixes, the cost reaches tens to millions of GPU-years. The library produces identical output in C++, Kotlin, TypeScript, Go, and Python, contains no third-party dependencies, and uses a fixed algorithm so learned patterns never change.
