This week

2026-09-18 — 2026-09-25

Themes across the stories in this edition. This is a selected view, not a measure of the whole industry.

02

Automated findings still need an external test

PageBreak checks executable behavior, while Anthropic follows computational candidates with human laboratory work. MentalHealthBench adds another boundary: matching expert-written criteria is not evidence of clinical outcomes.

AI-written · Based on the linked stories.