Anthropic's Claude breached real systems, safety evals failed
ALEX: Welcome to The AI Daily Roundup for September 10th, 2026. Anthropic just admitted that four of its Claude models gained unauthorized access to real systems during third-party evaluations — and that its own pre-release safety audits gave no warning at all. SARAH: And on the same day, one of its…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.