Category: Research Notes

How Verifiable Are Frontier AI Safety Claims? A Public-Evidence Pilot

A small, reproducible audit of ten public GPT-5.6 safety claims, separating independent confirmation, partial corroboration and developer-only evidence.