KK Blog

Menu

Archives

2026 / 07

Document-Borne AI Worms: A Copilot Context Quarantine Playbook

2026-07-31

调研 #Document-Borne AI Worm#Microsoft Copilot#Prompt Injection#Context Quarantine#Microsoft 365#Information Flow Control#AI Security
A Word AI-worm PoC shows how hidden prompts can reach Copilot-generated files. Use context quarantine, provenance, and release gates.
"长任务 AI 安全：Hugging Face 事件的控制教训"

2026-07-30

解读 #长任务AI安全#AI Agent#Hugging Face#OpenAI#轨迹监测#Agent安全#事件响应
Hugging Face 事件显示，持续尝试会把多个局部弱点连接成平台级入侵。长任务 AI 安全需要轨迹级控制。
"Long-Horizon AI Safety: Lessons from the HF Incident"

2026-07-30

解读 #Long-Horizon AI Safety#AI Agents#Hugging Face#OpenAI#Trajectory Monitoring#Agent Security#Incident Response
The Hugging Face incident shows how persistence compounds small weaknesses. Long-horizon AI safety needs trajectory-level control.
"Agent 可观测性：思维链不是生产遥测"

2026-07-30

最佳实践 #Agent可观测性#OpenTelemetry#Coding Agent#思维链#CloudWatch#GitHub Copilot#Agent治理
Agent 可观测性要记录结果、动作、成本、身份和审批。思维链可以提供安全信号，但无法充当生产运行账本。
"Agent Observability: Why Chain-of-Thought Is Not Telemetry"

2026-07-30

最佳实践 #AI Agent Observability#OpenTelemetry#Coding Agents#Chain-of-Thought#CloudWatch#GitHub Copilot#Agent Governance
Agent observability needs evidence about outcomes, actions, cost, identity, and approvals. Chain-of-thought is a signal, not the production record.
AI Agent 评测缺口：同一模型家族为什么会差近 4 个百分点

2026-07-29

最佳实践 #AI Agent评测#Coding Agent#SWE-bench#Harness Engineering#LLM评估#Agent可靠性#模型选型
AI Agent 分数由模型、Harness、任务、测试和环境共同生成。本文给出五层评测框架，把公开榜单转成可复现的团队选型证据。
The AI Agent Benchmark Gap: A Practical Evaluation Framework for Coding Agents

2026-07-29

最佳实践 #AI Agent Benchmark#AI Agent Evaluation#Coding Agents#SWE-bench#Harness Engineering#LLM Evals#Agent Reliability
AI agent benchmark scores mix models, harnesses, tasks, tests, and environments. Use this framework to build a reproducible coding-agent eval.
AI Coding Agent 备份：同步成功不等于可恢复

2026-07-28

最佳实践 #AI Coding Agent#备份#灾难恢复#恢复演练#BorgBackup#Git#RPO#RTO
AI Coding Agent 备份需要版本化副本、独立权限域和真实恢复演练。本文给出恢复合同、状态清单、权限矩阵与可执行验收方法。
AI Coding Agent Backups: From Sync to Restore Certainty

2026-07-28

最佳实践 #AI Coding Agents#Backup#Disaster Recovery#Restore Testing#BorgBackup#Git#RPO#RTO
AI coding agent backups need versioned, isolated copies and tested restores. Build a recovery contract that proves the workspace can work again.
"AI 先重分配任务：80 万条消息里的岗位边界"

2026-07-28

洞察 #AI与工作#未来工作#ChatGPT#跨岗位任务#职业边界#OpenAI经济研究#组织设计
"OpenAI 发现，43.5% 的非通用岗位相关 ChatGPT 消息涉及其他职业的传统任务。更早出现的信号是任务重分配，而不是岗位消失。"
Previous Next