August 14, 2026 · Observations on self-custody and AI agent security

1. The attack surface of hardware wallets isn’t limited to devices and private keys.

According to a Trezor announcement, their logistics provider ShipMonk was accessed without authorization: the names, email addresses, phone numbers, and shipping addresses of 11,742 customers were exposed, and another 1,947 people had their names, cities, and email addresses affected. The devices and Trezor’s own systems were not compromised; however, identity and logistics information can enable more precise phishing, fake customer service, and fake delivery risks.

2. An AI agent’s “hidden reasoning” may also become supply-chain data.

According to a republished research report by Decrypt, researchers decoded 315,320 pieces of reasoning from publicly shared AI agent conversation logs, finding data such as API keys, passwords, and personal information that were not originally visible in the output. The risk isn’t only that logs are made public: malicious instructions may also be hidden in inference blocks that security monitoring cannot easily observe, creating an invisible prompt injection.

3. Two things point to the same bottom line: key security can’t be answered only with “Is it encrypted?”

What we should ask instead is: who the data goes through, how long it is retained, who can see it, and whether the risk boundary can be clearly explained after an incident occurs. The former is logistics and identity information; the latter is the agent’s context and credentials. Neither is something that can be solved by a single technical control.

Disclosure: compiled by the CoWallet team. We build threshold ECDSA MPC wallets, so we have a position on self-custody and key security issues.