SIRF：面向工业内容风控的规则内化风险基础模型
SIRF: A Spec-Internalized Risk Foundation Model for Industrial Content Risk Control
AI 导读
研究者提出 SIRF（Spec-Internalized Risk Foundation Model），通过继续预训练将平台风控规则内化到模型权重，无需额外人工标注，规则合成采用 EntiGraph、MAGA 改写和账号级思维链。
正文 · 原文
Abstract:For industrial content risk control, the real deployment constraint is not average accuracy but how much risk can be auto-handled under high precision and second-level latency. We present SIRF (Spec-Internalized Risk Foundation Model), which internalizes a platform's complex policies, synthesized without additional human annotation via EntiGraph, MAGA rewriting and account-level chain-of-thought (CoT), into the weights via continued pretraining (CPT), so rules are applied at high precision under an ultra-low-latency, verdict-only deployment. A controlled same-source comparison (Qwen3-8B-SFT vs. SIRF-8B-SFT, identical policy injection and verdict-only output form, differing only in policy-grounded CPT) attributes the gain to internalization: SIRF-8B-SFT reaches 71.3% Black Recall@P95, +15.1pp over the baseline, using only ~70M CPT tokens without harming general ability, and among included, logprob-available models under this interface it matches or exceeds far larger systems. SIRF is deployed as a tree-model adjudication layer (20% more mis-penalized samples recovered) and transfers to a freezing scenario at low cost (~70% relative mis-penalization reduction).
| Comments: | 14 pages, 12 figures. Accepted at the Industry Track of EMNLP 2026 |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.11752 [cs.AI] |
| (or arXiv:2609.11752v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.11752 arXiv-issued DOI via DataCite |
Submission history
From: Suwan Wu [view email]
[v1]
Thu, 10 Sep 2026 16:08:54 UTC (2,922 KB)
[v2]
Fri, 30 Oct 2026 08:37:35 UTC (2,915 KB)
来源：arXiv cs.AI · arxiv.org