Seed-only August 21 X digest: Mythos-class privacy and 1 qualifying post from 1 author

Coverage note. This seed-only scan checked 86 public accounts and returned 669 posts. The August 21 Beijing window contained 127 posts; one original post from one author reached the 100-like threshold. The full @hwwaanng following list remains unavailable, so this edition covers the public seed sample, roughly 2.2% of the intended network.

Boris Cherny, who describes himself as Claude Code at Anthropic, posted at 4:34 a.m. Beijing time on August 21. His quote post carried the Bloomberg headline "Anthropic Plans to Change Data Retention Policy for Advanced AI" and said that Mythos-class models need additional safety measures, while enterprise customers need their own privacy and compliance rules. Cherny also wrote that customers can own and control their data, Anthropic retains none of it, and the change is coming this fall. The post had 2,199 likes and 123 retweets when the detail record was collected. 1

The post leaves several practical details open. Cherny does not name a specific model, give a date within the fall, or publish the retention terms. The announcement therefore gives enterprise readers a direction to watch: Anthropic is pairing a future model class with stronger safety requirements and a claimed customer-controlled data boundary.

Anthropic's fall announcement will need to answer four concrete questions: what qualifies as a Mythos-class model, which safety measures customers must meet, how customers can administer and retain their data, and how long Anthropic keeps any data that falls outside the "retains none" claim. Those details will show how much of the post is a forward-looking promise and how much is a usable enterprise policy.