Blog
Notes for people who have to be right.
How a Hong Kong practice should think about privacy, the model, and where the file actually goes.
Trust & product
Meet Makebell
Makebell is a Hong Kong company founded by legal professionals who had been working with AI since 2018. The system runs on the firm’s own machines.
1 November 2026
Trust & product
The largest firms are building AI they control
Latham bought the servers. Kirkland set aside $500 million for its own platform. The decision is the same one a smaller firm can make, without that budget.
31 October 2026
Trust & product
“We do not train on your data” still means the file left
A cloud vendor can promise not to train, and can even promise not to keep a copy. The advice was still sent. That is the gap Makebell is built to close.
30 October 2026
Trust & product
The model can draft. It can also be aimed at your systems.
Anthropic has published what capable models are already being used to do against computer systems. Where the file sits is a security decision, not only a drafting one.
28 October 2026
Trust & product
You no longer have to send the file away to get a strong model
DeepSeek, Qwen, GLM and others are now close to the closed models on a hard legal test. That is why a private system is no longer a compromise on quality.
24 October 2026
Trust & product
A seat price is not the cost of the work
Cloud looks cheaper because the comparison stops at the invoice. The work a firm cannot put on the tool, and the work staff already do in secret, are part of the price.
20 October 2026
Hong Kong & regulation
The risk is the file, the personal data, and the document you sign
Privilege, the PDPO, and a filing with the firm’s name on it are three different problems. A private system deals with the first two by never receiving the file. The third still needs a lawyer.
16 October 2026