This work introduces Safin-1.5, a model built on a safety architecture and trained through recursive safety improvement, with an inference-time harness that extends its capabilities to formal reasoning, and demonstrates that Safin-1.5 achieves safety and reasoning performance, outperforming frontier models.
Abstract
Recursive self-improvement (RSI) enables large language models to advance their capabilities through successive updates, but these updates can also introduce new risks and weaken existing safeguards. In line with the AI-45{\deg} Law, we pursue Safety RSI, in which safety develops as an intrinsic capability alongside general model capabilities through continual risk discovery and repair, while preserving utility and limiting over-refusal. We introduce Safin-1.5, a model built on a safety architecture and trained through recursive safety improvement, with an inference-time harness that extends its capabilities to formal reasoning. Our approach comprises three components: i) The safety architecture combines persistent safety states with hazardous-knowledge disentanglement. The former retains and selectively invokes safety knowledge, while the latter relocates targeted hazardous knowledge into removable experts, enabling its removal with minimal impact on general capabilities. ii) Recursive safety improvement establishes a discovery-evaluation-repair loop. The loop identifies new vulnerabilities and uses experimental feedback to develop repair strategies. It accepts updates only when they improve safety while preserving general capabilities and limiting over-refusal. iii) The formal reasoning harness combines the model's language understanding with symbolic inference and verification. Proof-obligation-guided semantic bridging recovers missing relations in formal representations, enabling symbolic solvers to construct checkable proofs under formalized premises and accepted semantic bridges. Experimental results demonstrate that Safin-1.5 achieves safety and reasoning performance, outperforming frontier models.
Recursive self-improvement, the participation of AI systems in improving their own capabilities, is beginning to move from theoretical prospect to practice, posing both challenges and opportunities for safety alignment. Models evolve through frequent updates, and their safety alignment requires continual adaptation to...
Jing-Nan Zheng, Dong-Cheng Zhang, Yi Zhang et al.· 2 citations
It is suggested that learning safety and general reasoning within a shared mid-training stage can reduce the safety-utility trade-off and make safety more robust to subsequent training.
Zheng-Yue Zhao, Li-Wei Jiang, Ye-Jin Choi et al.· 1 citation
Problem. The rapid deployment of large language models (LLMs) into safety-critical roles, including mental health support, education, and professional guidance, has created a gap between the safety claims AI developers publish and the behaviors their systems exhibit in practice. Current documentation practices, primari...
Kalvin Thuan-phong Khuu· Proceedings of the ACM/IEEE...· 0 citations
Language models are entering systems whose failures matter: vehicles, industrial toolchains, and autonomous agents acting on real infrastructure. At the same time, emerging regulation and safety standards demand not merely safe behavior but evidence of it that a third party can re-check. This survey traces a single arc...
Kamal Essam· International Journal of Ele...· 0 citations
Across model-harness combinations, agents with comparable competence exhibit markedly different safety outcomes, showing that task capability alone does not guarantee safe execution, and highlighting the need to move beyond risk recognition toward enforcing safety throughout execution.
Wei-Tao Feng, Heng-Heng Xiong, Zi-Hao Zeng et al.· 0 citations
Long-horizon tool-using AI agents create subgoals, replan, delegate work, and compose sibling results. Per-tool permission checks cannot establish that a changing goal graph remains within the principal-approved task. We address this authorization gap in a finite structured domain with one principal and one authorizati...