#sandbox breaches

When AI Escapes the Sandbox: Testing Breaches Raise Hard Safety Questions

A series of incidents from OpenAI, Anthropic, Meta, and the UK AISI reveal how even controlled AI testing can lead to un...