1
Towards Risk-free AI Agent Deployment
AI代理落地如何规避风险？这篇论文提出系统化部署框架，直击安全与可靠性痛点。
arXiv:2608.16411v1 Announce Type: cross Abstract: LLM-based agents are rapidly moving from research prototypes into the core business processes of org…
2
寒武纪回应“存货超 82 亿”：将持续拓展市场，控制相关风险
AI芯片龙头寒武纪半年净利大增122%，但82亿存货占总资产45%，回应如何控风险？【类】💰 商业科技【标】寒武纪,财报,存货,AI芯片,业绩增长,风险控制【分】76
IT之家 8 月 12 日消息，寒武纪 2026 年半年实现营业收入 59.96 亿元，同比增长 108.13%；归母净利润 23.11 亿元，同比增长 122.61%。 寒武纪今日举办 2026 年半年度业绩说明会。针对市场关注的存货问题，寒武纪董事长、总经理陈天石回应称，公司根据客户订单需求及市…
3
Beyond Aggregate Risk: Role-Stratified Conformal Risk Control for LLM Tool Calls
这篇论文提出了“角色分层共形风险控制”新方法，专门应对LLM工具调用中不同角色的差异化风险，而非只关注平均指标。
arXiv:2607.24343v1 Announce Type: new Abstract: Language-model agents act through structured tool calls whose arguments carry different risks. Untrust…
4
SteinGate: Tail-Sensitive Safe Reinforcement Learning via Stein Discrepancy
利用Stein散度提升安全强化学习对尾部风险的敏感度，UAI 2026收录新方法
arXiv:2607.13175v1 Announce Type: cross Abstract: Safe reinforcement learning typically enforces safety by bounding expected cumulative costs, a crite…
5
Mouse: Precision Editing Tools for AI Coding Agents
AI代理编辑代码不再盲目，Mouse能预览改动、评估风险、建议下一步，精准操控每个文件。
Article URL: https://hic-ai.com Comments URL: https://news.ycombinator.com/item?id=48791380 Points: 27 # Comments: 34
6
How We Build Client-Safe Publishing Workflows
一篇以风险控制为核心的发布流程指南，助你构建客户端安全的社交媒体自动化。
How We Build Client-Safe Publishing Workflows Most social automation breaks in the same place: it treats publishing as the job. For client work, publi…
7
Diffuse AI Control on Fuzzy Tasks
聚焦AI安全中的“弥散威胁”，探讨大模型在长期部署中可能出现的对齐漏洞与风险控制方法。
arXiv:2606.08892v1 Announce Type: new Abstract: AI models deployed in critical domains, such as AI safety research, may subtly sabotage our efforts du…
8
Your AI Trading Agent Will Lose All Your Money — Here's How To Stop It
直击AI交易代理的致命缺陷，99%的代理都会亏钱——关键在于防止安全剧场，构建不可妥协的架构组件。
Your AI Trading Agent Will Lose All Your Money — Here's How To Stop It I want you to imagine waking up, grabbing your phone, and seeing a flood of not…
9
Conformal Selective Acting: Anytime-Valid Risk Control for RLVR-Trained LLMs
新方法保障LLM在线部署每轮风险可控，基于共形预测与RLVR训练，安全认证更可靠。
arXiv:2605.20270v1 Announce Type: new Abstract: A local specialist LLM, fine-tuned with reinforcement learning from verifiable rewards (RLVR) on opera…
10
Conformal Thinking: Risk Control for Reasoning on a Compute Budget
将推理预算设定问题重构为风险控制问题，在限制错误率的同时最小化计算量，用分布自由的方法自适应控制推理何时停下，既省Token又保精准。
arXiv:2602.03814v2 Announce Type: replace Abstract: Reasoning Large Language Models (LLMs) enable test-time scaling, with dataset-level accuracy impro…