Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 12 papers Another 255 papers do not have category tags yet and are not included in filtered results.
“
About this collection Collection in progress We are building a large AI safety paper collection to help researchers find work related to their interests. Papers are collected automatically andfiltered only for relevance to AI safety , without judging quality, novelty or reading value. The default “Annotated first” shows papers with tags or published paper analyses first. Each group is ordered by original publication date, newest first, with undated papers last. “Newest first” uses publication dates only. This order is not a quality ranking.
One-line summaries, tags and paper analyses are prepared by AI from the papers. Language models may have systematic biases and produce hallucinations.We plan to carry out human annotation and review to reduce errors in this content.
Proposes RSI-Anything, which uses a world model to select candidates and retain policy updates.
Published Oct 10
Evaluated models ReasonFlux-Coder-7B , 4B policy , W1 Abstract RSI-Anything integrates world models into recursive self-improvement through prediction, feedback, and retention checks, and reports gains across multiple domains.RSI-Anything is a cross-domain framework: it uses world model predictions to decide which candidates to take external feedback for, then uses the feedback to decide whether the policy, workflow, or predictor is updated, and uses checks to decide which pair is retained. The difficulty to address is the limited interaction budget, and the fact that once the policy changes, the candidate distribution faced by the predictor also changes. The authors argue that improving only one side does not guarantee that the assembled system will be better.
提出 Mid-Harness，在执行前验证并筛选终端智能体的候选动作
Published Sep 30
Evaluated models TMAX-9B , TMAX-4B , TMAX-27B and others6 Abstract Mid-Harness 在执行前选择候选动作。Mid-Harness 在不改生成器和执行 harness 的前提下，于二者之间做动作级测试时计算，用来研究终端智能体里“多采样、再验证”何时提高轨迹成功。
提出 PatchHolmes，用混合检索和列表式智能体选出 CVE 修复提交
Published Sep 30
Evaluated models Qwen3-235B , Qwen3-Coder-30B , gpt-oss-120B and others4 Abstract PatchHolmes 用混合检索和列表式四工具智能体做补丁检索，增益主要来自联合阅读候选而非更换骨干。PatchHolmes 是面向补丁检索的两阶段系统：在本地 Git 仓库上用冻结开源权重模型，为一个已知 CVE 找出修复提交，不做按仓库微调，也不使用付费外部搜索 API。
提出 SelfSearch，用自我修改记录无奖励搜索改进编码智能体
Published Sep 30
Evaluated models gpt-5.6-luna , deepseek-v4-flash-0731 提出 TAIP 持续保障，在策略或证据变化后重算仓库审计交付门裁决
Published Sep 28 Abstract TAIP 把策略、证据和执行分开，在保留的 RepoAudit 证据上于 5s 预算内重算合并门姿态。TAIP 是放在未修改 RepoAudit 之外的保证层，用来在软件交付门上重算智能体安全控制的当前姿态。
提出 Opera 口头批评框架，跟踪长程编码 Agent 的执行纠正
Published Sep 28
Evaluated models Qwen3.5-9B (thinking) , Muse-glimmer-30B (high) , Qwen3.8-27B (xhigh) and others4 Abstract Opera 用可审计的持久纠正笔记提升长程编码智能体，并可用引导轨迹做近似 on-policy 微调。Opera 是给长程编码智能体用的口头批评框架：每次纠正是一条要跟到解决的持久笔记，而不是一次性评语。
勾勒科学智能体经济以协调验证、信用与安全
Published Sep 26 Abstract 作者主张以智能体经济管理自主科学发现的验证资源、信用、问责与安全，而非只提升推理。作者把高自主 AI scientist 的规模化协调写成经济与制度问题：假设生成变便宜之后，要分配的是稀缺验证资源，并处理信用、问责和恶意使用。
提出双层系统 AIDE2，使研究智能体自改进 Harness 并对比人类研发基线
Published Sep 22
Evaluated models gemini 3 flash , gpt-5.6-sol , fable 5 Abstract 提出双层递归自改进系统 AIDE2，在固定预算下演化 Harness 代码，多项留出基准超越人类基线并降低奖励黑客。
提出 Djinnlang，使人只写无歧义规格，由编译器内 LLM 补实现并经 Dafny 核验
Published Sep 21
Evaluated models Claude Fable 5.1 (Claude Code) , GPT 6 Astra (Codex) Abstract 无歧义规格加验证器，使 LLM 成为可丢弃代码的编译阶段。Djinnlang 让程序员只写规格，把 LLM 放进编译器，并用 Dafny 验证器加上无歧义约束钉住语义。
提出双缺口框架，统一智能体软件工程中的奖励投机与幻觉
Published Sep 11 Abstract 双缺口解释内环为何会放行部署后不可接受的实现，外环用现实证据持续修订 R、M 与 E。Reality Is the Final Verifier 用 two-gap framework 解释智能体软件工程的主要失败，并主张用外层保证环持续收窄缺口，而不是试图在部署前证明缺口已关闭。
提出 AgentRewind，支持长程 Agent 回退对齐检查点并继续执行
Published Aug 14
Evaluated models GPT-5.4 , GPT-5.4 mini , Qwen3.7-Max and others7 Abstract 作者主张防护优先于自主性，用三元框架把输出安全扩展到行为与环境反馈。这是一篇关于 LLM 驱动 AI scientists 的 perspective：自主科研能力增强时，误用和意外后果缺少科学情境下的系统防护。
You have reached the end