AI this week: where your data goes, the mystery model unmasked, AI for lawyers

OpenAI and Anthropic’s data-retention moves, “Ox Alpha” revealed as GLM-5.3-Flash, and Thomson Reuters builds its own model.

September 4, 2026 · By Bohol AI

This week was about trust: what AI companies keep from your conversations, who was behind the internet’s most-used mystery model, and why some companies want models trained only on their own data.

What happens to your data

Business customers of OpenAI and Anthropic got new answers about data retention. OpenAI previewed zero data retention for its frontier models. Anthropic confirmed retention practices for its most capable models after details leaked. Technical details remain thin for both.

Three questions are worth asking any AI service before you share private information: where is it stored, how long is it kept, and what is it used for, such as training or human review?

“Ox Alpha” was GLM-5.3-Flash

For more than a week, the most-used model on OpenRouter was an anonymous preview called Ox Alpha. Z.ai revealed it as GLM-5.3-Flash, its first vision model in months, said the free preview ran entirely on Chinese-made chips, and released the weights for anyone to download.

Thomson: a model for law, tax and finance

Thomson Reuters launched Thomson, its own model family built on a Qwen model and retrained with the company’s data for law, business, tax, finance and news, aiming for more accurate answers than general models in those fields.

Research: simpler safety monitoring

Researchers found that checking a single safety score during generation catches harmful or incorrect output nearly as well as analysing a whole series of scores, which makes monitoring cheaper.

What it means for Bohol

Before pasting client, student or patient information into any AI tool, ask the three data questions above. If you can’t get answers, don’t paste it.
Specialised models for law and finance suggest a future where local firms use AI trained on their own records, kept private.

Sources

All articles