Hardware Wallet Coin Isolation Bypass
TLDR: Security researcher monokh discovered a flaw in the isolation of the transaction logic between coins. It allows coins from one cryptocurrency, e.g. bitcoin, to be spent by the transaction flow of another, e.g. Litecoin. I found two other hardware wallets that are vulnerable to this exploit, Keepkey and Coldcard. To this day, Keepkey has not patched the issue.
Security and bitcoin researcher monokh published an article on his blog on August 04 2020 titled “Ledger App Isolation Bypass”. It describes a flaw in the isolation between the various apps that can be installed on a Ledger. From monokh’s post:
It was discovered that for Bitcoin and Bitcoin forks, the device exposes its functions for any of the assets. In other words, having unlocked the Litecoin app, you will receive a confirmation request for a Bitcoin transfer while the interface presents it as a transfer of Litecoins to a Litecoin address. Accepting the confirmation produces a fully valid signed Bitcoin (mainnet) transaction.
Essentially keys intended for one cryptocurrency could be accessed by another, allowing coins with the same signature format to be cross-spent by each other. This breaks the most fundamental assumption of a hardware wallet: You sign what you see.
A few days later Trezor acknowledged another security researcher discovering a very similar vulnerability in the Trezor One. Again the user could be tricked into signing a Bitcoin transaction while using the workflow of an Altcoin transaction. Trezor describes this in a blog post.
Hardware wallet vulnerabilities across vendors
I started tracking vulnerabilities across most hardware wallet vendors a year ago in my list of hardware wallet hacks. I observed that vendors tend to repeat their mistakes a lot. These range from non-constant time PIN-checks to insufficient checks on change addresses and signature replay attacks.
With the two largest hardware wallet vendors, Trezor and Ledger, making the same mistake, I suspected other vendors repeating their mistake as well. I also saw many posts on Twitter and Reddit blaming the vulnerability on the altcoins, concluding a hardware wallet not supporting altcoins is not vulnerable to the attack. The owner of the Coldcard parent company coinkite even tweeting in response to the vulnerability:
Shitcoins are a hardware wallet liability. Coldcard unaffected 😉— UNCORELETED OPTIMISM (@nvk) August 4, 2020
However, the vulnerability also extends to Bitcoin testnet. The user could be visually verifying a testnet transaction, while actually signing mainnet.
Wallets with monokh’s vulnerability
In the following weeks, I tested monokh’s exploit on the Cobo, Keepkey, Coldcard and BitBox02 wallet. Both Cobo and the BitBox02 were not vulnerable. Keepkey is based on the Trezor One, so it was no surprise that it had no checks enforcing the isolation. Coldcard also had no checks enforcing isolation between Bitcoin testnet and mainnet. Independently, benma, my ex-colleague from my days working on the BitBox02, had found and disclosed the vulnerability to Coldcard already. Benma also publicly disclosed the issue in a blog post without Coldcard having patched the issue yet. This left me to contact Keepkey’s, or rather Shapshift’s, security team on August 20.
Keepkey was unable to deliver a fix within a typical 90 day disclosure period. Though they were responsive to my messages, they were unable to commit to a patch timeline. Since the outline of the vulnerability is public knowledge already, I am now publicly disclosing the vulnerability.
Keepkey recently fixed a host of exploits in a recent release, which they detail in a blog post. I am disappointed that the Keepkey team is not committed to patching vulnerabilities in their hardware wallet within reasonable time. I therefore strongly recommend against buying one or promoting its use.
Coldcard provided a fix to the issue in their version 3.2.0 release by moving the testnet screen behind an additional “Danger Zone” warning screen, which they explain in another blog post. Since this extra step protects users from being duped into the testnet workflow, I consider this a fix for the issue.