While organizing documents, I found a file about HT burn statistics. Unexpectedly, a few years later, all the exchanges were burned.

Cloudflare's agent wallet. Didn't OKX already make one a long time ago? OK should probably also create one first to reserve the ID and build some hype.

A sign that the computer is getting old is that my nearly six-year-old M1 Max MacBook Pro has started to heat up.

Clodcard official emphasized "using the latest models such as Kimi K3, Claude Fable, Codex 5.6, no vulnerabilities were detected afterward," intending to illustrate the limitations of AI and trying to find an angle to shirk their own responsibility. But logically, this actually contains a profound paradox: 1. The paradox of time and retrospective review: Using current AI to test firmware code that was released long ago is itself a temporal inversion. No matter how you review it afterward, you cannot fix or change the established facts of risks caused by the past releases. 2. The paradox of generation and review: When the code was originally written, there was no AI as powerful as today. If top-tier AI had been used at the development source back then for full-stack collaboration and generation, with its rigorous analysis of overall architecture and submodule boundary conditions, it most likely would never have produced such submodule interaction errors from the start. Using "current AI did not detect vulnerabilities in old code from the past" to negate AI ignores the essential difference between AI's "source specification generation" and "retrospective static review."

There are so many AI models that you can't use them all. Besides ChatGPT and Claude, Grok can be used continuously, Gemini/Antigravity can handle light tasks. If that's still not enough, there's Deepseek. There's basically no chance to get to the top-tier priced Kimi and GLM.

The address of the first wave of attacks, bc1qnk4zh9qcnap2mycp56qjrgza3cc8ylrh8fecp0: about 32.45 BTC remains to be collected as residue. There are also two inscriptions in the address, which are two btc domains. Randomly opened a stolen address that hasn't moved since 2021. All are small transactions, accumulated bit by bit, transferred from exchanges to the wallet. After many years, 0.2 BTC was accumulated and then completely looted.

Also be careful of scam emails in your inbox.

If the seed generated by the hardware wallet and the passphrase are truly random and independent, then their combined entropy is approximately additive. The reason is that the number of candidate combinations multiplies, while entropy is the base-2 logarithm of the number of candidates. Currently, there are no public reports of funds being stolen from Coldcard wallets with a strong BIP-39 passphrase (password phrase/25th word) set.

A harsh truth is: coldcard is also open source, Source code is public ✅ Can be self-compiled ✅ Reproducible Build ✅ Official firmware signature verification ✅ Coldcard basically meets all four points. But except for hackers, no one verifies the code security. Until a major incident occurs.

Recently, I've been frequently seeing videos of Zou Shiming and her financial master. It doesn't feel like acting. Her results match her understanding.