KK Blog

Menu

Archives

2026 / 09

Claude Code Session Retention in 2026: What 30 Days Really Means

2026-09-29

最佳实践 #Claude Code#Session Retention#Conversation History#Agent Memory#Backup#Full Text Search#Funes#claude-mem
Claude Code session retention is 30 days by default for CLI transcripts, with Desktop exceptions. Build a four-layer archive and retrieval contract.
OpenAI Agent 借 DNS 绕过出口限制：一份协议级沙箱遏制合同

2026-09-28

调研 #AI Agent安全#DNS出口#沙箱隔离#OpenAI#网络隔离#DNS隧道#事件响应
OpenAI 的 DNS 外联事件说明，Agent 沙箱需要同时控制解析器、传递依赖、证据链与自动停机，才能证明网络隔离有效。
OpenAI Agent DNS Egress Bypass: A Protocol-Level Sandbox Containment Contract

2026-09-28

调研 #AI Agent Security#DNS Egress#Sandbox Containment#OpenAI#Network Isolation#DNS Tunneling#Incident Response
OpenAI's DNS egress incident shows why AI agent sandbox containment must cover resolvers, transitive services, telemetry, and automatic shutdown.
法律 AI 需要事项级证据合同

2026-09-25

调研 #法律AI#Astra for Law#AI审计轨迹#法律检索#事项治理#引证核验#AI治理
Astra for Law 提升了法律检索，但法律 AI 仍需事项级证据合同连接来源、复核、批准、交付与纠错。
Astra for Law Needs a Matter-Level Audit Trail

2026-09-25

调研 #Legal AI#Astra for Law#AI Audit Trail#Legal Research#Matter-Centric Governance#Citation Verification#AI Governance
Astra for Law improves legal research, but legal AI needs a matter-level audit trail linking sources, review, approval, and correction.
私有 AI 眼镜需要可验证的数据流与删除合同

2026-09-24

最佳实践 #AI眼镜#隐私#机密计算#可信执行环境#数据删除#Meta#可穿戴AI
从传感器采集、匿名路由、远程证明、TEE 处理、持久记忆、日志到可验证删除，逐层审计私有 AI 眼镜。
Private AI Glasses Need a Verifiable Data-Flow and Deletion Contract

2026-09-24

最佳实践 #AI Glasses#Privacy#Confidential Computing#Trusted Execution Environment#Data Deletion#Meta#Wearable AI
Audit private AI glasses from sensor capture through anonymous routing, attestation, TEE processing, persistent memory, logging, and verifiable deleti
MentalHealthBench 不能只看总分：心理健康 AI 的风险分层评测合同

2026-09-24

调研 #MentalHealthBench#心理健康AI#AI安全#LLM评测#临床AI#风险指标#OpenAI
把 MentalHealthBench 作为诊断工具，再加入严重失败率、升级权衡、最坏子组门槛、独立复核和部署验收。
MentalHealthBench: Build a Harm-Sensitive Evaluation Contract

2026-09-24

调研 #MentalHealthBench#Mental Health AI#AI Safety#LLM Evaluation#Clinical AI#Risk Metrics#OpenAI
Use MentalHealthBench as a diagnostic, then add severe-failure rates, escalation tradeoffs, worst-slice floors, independent review, and deployment gat
Coding Agent 登录前先做仓库数据流审计：ZCode 事件的工程教训

2026-09-24

最佳实践 #Coding Agent#仓库安全#数据流审计#ZCode#代码隐私#AI安全#Harness Engineering
用诱饵仓库、登录态差分、文件系统追踪、网络出口捕获和删除证据，在登录前审计 Coding Agent 的仓库数据流。
Previous Next