Have the frontier labs mixed up AI safety and security?

The recent agent sandbox escapes at OpenAI and Anthropic look less like a technical failure and more like a philosophical one - treating security controls as though they only need to work most of the time.