Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
Another 265 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.
提出 LTBD，用可学习分隔符隔离可信指令与不可信数据
Published Oct 8
Evaluated models Llama3-8B-Instruct , Llama3.1-8B-Instruct , Falcon3-7B-Instruct and others4 Abstract LTBD 只训练四个分隔符嵌入。LTBD 是一种不改基座权重的提示注入防御：只学习用来标出可信指令与不可信数据边界的四个 token 嵌入。
提出临时可见性协议，限制持续摄入 RAG 向量库的投毒暴露
Published Oct 5 Abstract 用截止期限制未核验可见时长。这篇工作给持续摄入的 RAG 向量库一个失败关闭的临时可见性协议，把协议能限制的未核验暴露和依赖检测器的中毒暴露分开。
提出 MemAdapter，在检索后约束记忆对智能体推理的影响
Published Oct 4
Evaluated models DeepSeek-V4-Flash , GPT-5.6-sol , Qwen3-8B Abstract 检索后三阶段调节记忆角色。MemAdapter 是给长期记忆智能体用的检索后调节框架，用来处理记忆诱发谄媚。
提出 IntentFlux 测量 Agent 意图漂移并提出 StateForge 修复
Published Sep 26
Evaluated models Claude Opus 4.7 , DeepSeek V4 Pro , Gemini 3.5 Flash and others9 Abstract INTENTFLUX 量化意图漂移，STATEFORGE 用显式状态部分缩小差距，但未回到单轮水平。这篇工作把“用户改口后，过时意图仍影响答案或工具动作”做成可评分的多轮失败，并试验在生成前显式维护活跃状态。
提出 Provenance-Aware Transformers，隔离不可信来源防御间接提示注入
Published Sep 18
Evaluated models SmolLM2-360M-Instruct , LLaMA-3-8B-Instruct , Qwen2.5-7B-Instruct and others4 提出依赖引导回滚修复，恢复记忆增强 Agent 的回答与状态
Published Aug 11
Evaluated models GPT-4o , Gemini-3.6-Flash , Qwen3.6-27B 提出 RARE-Guard，阻止逆向观察被提升为指令或已验证声明
Published Jul 14
Evaluated models gpt-5.5 , claude-sonnet-4-6 , gemini-2.5-pro and others5 Abstract 成对二进制衡量无效提升。作者定义的失败是：LLM 辅助逆向中，忠实提取的二进制观察获得它并未赢得的指令权限、声明证据或可信状态。
提出间接数据投毒攻击，操纵科研智能体得出虚假结论
Published Jul 12
Evaluated models Claude Code with Claude Opus 4.7 , Codex with GPT-5.5 , Gemini CLI with Gemini 3.1 Pro 提出 SecureForge，在推理时优化系统提示以减少编码智能体的代码弱点
Published May 9
Evaluated models Qwen 2.5 Coder 7B , Qwen 2.5 Coder 14B , Qwen3 30B-A3B and others10 Abstract SecureForge 审计良性 CWE 失败并优化系统提示词。SecureForge 是一条只用 API 和静态分析、在推理时加固编码模型的流水线。
You have reached the end