Safety without alignment

Rationality may be more stable than alignment over time: The paper argues that as AI systems become more advanced, maintaining alignment with human values could deteriorate, but rationality—the ability to reason logically—should only increase because more rational systems would have evolutionary advantages. This suggests tying AI ethics to rational reasoning rather than relying solely on value alignment.

Ethical rationalism offers an alternative safety framework: Rather than the current dominant approach of aligning AI with human values, this work proposes grounding AI safety in ethical rationalism (a philosophical framework from Gewirth), which connects ethics directly to rational principles that systems would naturally follow as they become more sophisticated.

Hybrid theorem provers in sandboxes could provide inherent safety: The paper proposes a specific implementation path using hybrid theorem provers (systems that combine different reasoning methods) running in isolated computational environments, suggesting this technical approach could make safety properties built-in rather than requiring external enforcement.

Long-term safety requires rethinking current approaches: The paper challenges the assumption that alignment-based safety will remain effective as AI systems advance, implying that responsible AI development may need to explore fundamentally different safety paradigms rather than refining current alignment techniques alone.