What's this about?
Thomson Reuters, which owns decades of legal, tax and news data, announced it is launching its own frontier model built on those proprietary assets rather than licensing someone else's.
What exactly is a "data moat"?
When every lab can train a competent general model, the thing you cannot copy is exclusive data. Thomson Reuters owns case law, tax rulings and editorial archives that are not freely scrapeable. The bet is that a model trained on that corpus beats a better general model on legal and tax work, because the advantage is in the data rather than the architecture.
What's the community saying?
The discussion is modest and mostly focused on whether "frontier model" is the right word for a domain-specific system, and on whether owning the data is enough when the underlying model technology keeps commoditising.