Is Reasoning Capability Enough for Safety in Long-Context Language Models?

It is suggested that safety does not automatically scale with reasoning capability, especially under long-context inference, as context length increases and safety alignment consistently degrades.