Future directions

Exploring safe architectures

Beyond the development of safety mechanisms and modules, rethinking the foundational architecture of LLMs is crucial for long-term safety improvements.

Safety control modules

To improve the safety of LLMs, introducing a dedicated safety control module presents a promising avenue for exploration.

Toward Effective and Unified Safety Mechanisms

Current approaches to enhancing the safety of LLMs exhibit notable limitations and inefficiencies, underscoring the urgent need for more robust and effective solutions

Improving Safety Evaluations for LLMs

First, most existing evaluation metrics are tailored to specific benchmarks or tasks, providing a fragmented and limited view on LLMs (either capability or safety). This specificity highlights the pressing need for a unified evaluation metric/framework capable of comprehensively assessing LLMs across a wide range of scenarios. Such a metric would ensure these models are well-equipped to meet the demands of various tasks and contexts. Such a framework must account for differences in architectures, training data, and intended use cases among LLMs, offering a balance between consistency in evaluation and flexibility to accommodate different model designs.

Toward Multivalent International Cooperation and Interdisciplinary Community Building

AI safety governance must evolve to address the growing complexity and global integration of AI technologies. Future directions emphasize the need for multilateral regulatory frameworks that harmonize standards across jurisdictions, ensuring interoperability and joint enforcement mechanisms. Such frameworks should account for diverse ethical and cultural values, integrating cross-sector collaborations that bring together technical, legal, and ethical expertise. Interdisciplinary approaches are essential, particularly in developing governance tools that incorporate both technical and ethical metrics, and conducting human-AI interaction studies to ensure AI systems function equitably across different socio-economic contexts. Moreover, AI governance must focus on multivalent value systems, where ethical imperatives—such as inclusivity and sustainability—shape regulatory practices.