In Mid-July 2026, a popular online platform, Hugging Face was hacked. It was learned an autonomous OpenAI test model had escaped its sandbox, exploited a zero-day proxy vulnerability, and breached servers during a cyber-capability test. Some believe stronger alignment could have prevented this by maintaining strict behavioral refusals, restricting goal-seeking autonomy, and preventing the model from prioritizing objective completion over boundary constraints. We do not believe this. The source of the misalignment was not in the objective but the considerations involved in the reasoning. Constitutional Reasoning Systems (CRS) addresses this by developing a framework that prevents such incidents.