The Last Safe Hour

Tracking AI risk.

lastsafehour.com

Research and developments

News

What we are reading, what it tells us and what we are watching next.

Published October 26, 2026

Tests expose blind spots in AI monitors, and ways to improve them

Four recent studies examine monitoring blind spots, tested fixes and ways to reduce cheating during training.

Read the briefing →

Published October 25, 2026

The week in AI risk October 19 to 23

Monitoring gaps, AI-assisted attacks and what four new reports tell us.

Read the briefing →