FINDING HIGH-SEVERITY SECURITY ISSUES WITH PUBLICLY AVAILABLE MODELS

Summary from the TLDR AI newsletter

Ramp pointed roughly 10,000 Inspect coding-agent sessions at its backend in an 8-hour run with a minimal "find security issues" prompt.