Safin-1: Safety from Within through Memory-Native State Evolution
- 类型：arxiv
- 标识：2609.00092
- 链接：http://arxiv.org/abs/2609.00092v1
- 主分类：risk
- 形态：method
- TLDR：Long-horizon complex tasks require foundation models to accumulate information, maintain internal states, and adapt over extended interactions. Safety should be an intrinsic property of the model itself, rather than a behavioral constraint relying solely on external safeguards or post-hoc alignment such as supervised fine-tuning. This motivates Safety from Within, where safety-relevant capabilities are represented and invoked through the model's native computation. We present Safin-1, a family of foundation models realizing this principle through memory routing and state evolution. Safin-1 is
- 待LLM分类：否
- 来源文件：
- /inbox/tom/_candidates/2026-09-02-agent-rag-longcontext-candidates.json