Trezor Model T flaw let malware divert transaction change to a hidden multisig address

On 18 November 2019 Marko Bencun of Shift Cryptosecurity, the company behind the BitBox hardware wallet, published details of a flaw in how the Trezor Model T verified the change output of a Bitcoin transaction. He states that malware on the user’s computer could silently replace the change with an output controlled by the attacker, taking all funds spent in the transaction except the amount sent. He reports that the Trezor One was not affected and that he performed the attack on Bitcoin testnet through a modified Electrum.

According to the write-up, the Model T accepted a multisig change output without showing it if its configuration matched that of all multisig inputs, but it did not check that all inputs were of the same type. An attacker could therefore add one small 1-of-2 multisig input shared with the attacker to an ordinary single-signature transaction and route the change to that 1-of-2 address, which the attacker alone can then spend; adjusting the change amount keeps the displayed fee unchanged. Trezor’s security portal describes the issue as a time-of-check/time-of-use bug that let a crafted multisig transaction include an attacker’s change output the user never confirmed.

Trezor’s security portal lists the issue as reported by Marko Bencun on 1 October 2019, affecting the Trezor Model T and fixed in firmware 2.1.8. Bencun states the fix was released on 6 November 2019, that SatoshiLabs notified users and paid a bug bounty. In a separate article on multisig pitfalls, Shift Crypto calls it a critical remote attack and describes related issues it reported to Coldcard (fixed in Coldcard firmware 3.0.2) and, on 4 November 2019, to Ledger, which responded that the behaviour was expected.