How did attackers steal €60.7 million from cold wallets without touching the devices?

Most attacks in the cryptocurrency world are based on hacking into online exchanges, exploiting smart contracts, or using tricks to obtain keys from a computer. Cold hardware wallets are considered the gold standard of protection precisely because they don't expose the keys to the internet. However, the latest incident involved a completely different scenario. Instead of stealing an existing key, the perpetrator simply recreated the keys themselves.

Researchers from Galaxy Research found that a total of 1,082.65 bitcoins were stolen. The funds were transferred in six blocks with pauses in between, suggesting a batch transmission of transactions. The stolen funds are now resting on four separate web addresses, with initial reports only detecting one of them, making the final damage estimate much higher than initially thought.

The core of the problem lies in the way the device generates random numbers, from which it creates a so-called seed. The Coldcard device software should use a dedicated hardware component to generate randomness. However, due to an internal error in the code compilation, the device skipped this crucial step. The support library continued to function smoothly, only checking whether the setting existed, not whether it was actually enabled.

Instead of hardware randomness, the system switched to a simple software replacement based on the chip serial number and the clock data. Since the serial number is fixed factory information, and the clock operation can be estimated or measured with relative accuracy, the set of all possible keys was reduced. Instead of an unimaginably large combination of random numbers, the set was reduced to about four billion possibilities for the Mk4, Q and Mk5 models, with the older Mk2 and Mk3 models also affected.

Although four billion sounds like a lot, it is an extremely small number for modern computers. The attacker was able to calculate all possible combinations on his equipment, generate addresses from them, and compare them with the publicly available data chain. The owner's wallet was not even needed. The scale of the attack is also confirmed by the fact that bitcoin transfer addresses were divided into three different technological formats, which indicates a systematic investigation of all possibilities.

Despite careful planning, the attackers made a mistake. According to Block, they used a paid account with a well-known blockchain data provider while investigating the network. The provider's logs accurately recorded the time and sequence of requests, and the details have already been handed over to law enforcement.

While Coinkite says its newer devices are safe and that the issue only affects the Mk3 model, Block researchers warn of a wider range of models being affected. Until the matter is fully understood, caution is advised. Anyone who has created a “seed” on affected equipment should assume the worst-case scenario and move their funds safely as soon as possible.

Research from Anthropic also warns that artificial intelligence is increasingly being used to crack security algorithms. This only further proves that storing the key securely is only half the battle. The other half is ensuring that the key cannot be automatically generated.