{ "report_date": "2026-05-17", "title": "AI 日报 2026-05-17", "summary": "今天可回源的新发布偏少，主体信息收敛为 5 条：ThreatConnect 8.0 把威胁情报工作台加入 Agentic AI，AWS Partner Central agents 用自然语言和 MCP 推进机会创建，Firecrawl v2.10 强化 agent 数据采集控制，xAI 发布 Grok Build 终端 coding agent early beta，OpenAI 披露 ChatGPT 在敏感对话中的跨轮上下文安全更新。本轮显式扩展到 2026-05-14，以补足 48 小时内不足 5 条的高质量一手来源，并把已在 2026-05-16 覆盖的条目降级为社区线索。", "main_items": [ { "title": "ThreatConnect 8.0 发布 Agentic Threat Intelligence Platform", "event_date": "2026-05-16", "url": "https://knowledge.threatconnect.com/docs/8-0-release-notes", "source": "ThreatConnect Release Notes", "tier": "T0", "entities": [ "ThreatConnect", "Agentic AI", "Threat Intelligence", "STRIDE", "Playbooks" ], "summary": "ThreatConnect 8.0 release notes 显示，该版本加入 Agentic AI beta，把威胁情报查询、STRIDE 威胁建模、情报报告生成和 Intelligence Requirement 创建做成四类可从 Ask AI 入口调用的代理能力。", "bullets": [ "四个代理都围绕已有 ThreatConnect 情报和 workflow 数据工作，官方说明它们基于 playbook architecture，并在 LLM reasoning layer 之外接入实例内情报和流程数据。", "ThreatConnect Query Agent 会给出 source links 供继续核验；STRIDE Threat Modeling Agent 可从 Ask AI 或 Threat Actor Profile 详情页生成结构化报告。", "8.0 同时加入 Playbook Routes 和 Parameter Sets，用更少节点表达分支逻辑并复用 app 参数，这说明安全运营平台的 AI 功能正和既有自动化编排层绑定，而不是只做一个聊天侧栏。" ], "image_urls": [], "importance": "major", "editorial_category": "ai_industry" }, { "title": "AWS Partner Central agents 支持用自然语言创建机会并提供 MCP 入口", "event_date": "2026-05-15", "url": "https://aws.amazon.com/about-aws/whats-new/2026/05/aws-partner-central-agents-oppo/", "source": "AWS What's New", "tier": "T0", "entities": [ "AWS", "AWS Partner Central", "Amazon Bedrock AgentCore", "Amazon Q", "MCP" ], "summary": "AWS 宣布 Partner Central agents 现在可通过自然语言对话加速 opportunity creation；该能力基于 Amazon Bedrock AgentCore，面向合作伙伴的 co-sell pipeline、deal hygiene 和 funding 机会识别。", "bullets": [ "合作伙伴可以用自然语言描述交易、上传 meeting notes、proposals、call transcripts，或 clone 现有 opportunity；代理会抽取信息、补充 customer details 并建议修正缺失上下文或字段。", "入口同时覆盖 AWS Console 中的 Amazon Q chat 和可编程 Model Context Protocol，因此销售团队可以从既有工具中触发机会创建，而不是只能进入 AWS Console 表单。", "该功能已在所有 commercial AWS Regions 提供，说明 Bedrock AgentCore 正被 AWS 用于内部/伙伴业务流程，而不仅是通用 agent 开发样例。" ], "image_urls": [], "importance": "major", "editorial_category": "ai_industry" }, { "title": "Firecrawl v2.10 合并发布 parse、Lockdown、Question/Highlights 和新 SDK", "event_date": "2026-05-15", "url": "https://www.firecrawl.dev/changelog", "source": "Firecrawl Changelog", "tier": "T0", "entities": [ "Firecrawl", "v2.10", "Parse", "Lockdown Mode", "MCP" ], "summary": "Firecrawl v2.10 changelog 把近几周的数据提取和安全功能合并到一个版本：`/parse`、Lockdown Mode、Question format、Highlights format，以及 Go、Ruby、PHP、.NET 官方 SDK。", "bullets": [ "`/parse` 可把 PDF、Word、spreadsheet 等文件转成 LLM-ready Markdown、JSON 或 summary，并标注文件上限为 50 MB，适合 RAG ingestion 和 agent 文档读取。", "Lockdown Mode 允许 `/scrape` 只从 Firecrawl index 返回结果，不发起 outbound requests，并默认 zero data retention；该能力覆盖 API、CLI 和 MCP。", "Question 和 Highlights format 的重点是上下文压缩：前者返回 page-grounded answer，后者返回匹配 query 的原句、代码块和表格行；对 agent loop 来说，这是把网页读取从“整页塞上下文”改成可控提取。" ], "image_urls": [], "importance": "major", "editorial_category": "ai_industry" }, { "title": "xAI 发布 Grok Build 终端 coding agent early beta", "event_date": "2026-05-14", "url": "https://x.ai/news/grok-build-cli", "source": "xAI News", "tier": "T0", "entities": [ "xAI", "Grok Build", "Coding agent", "CLI", "MCP" ], "summary": "xAI 宣布 Grok Build early beta，面向 SuperGrok Heavy subscribers 提供一个直接在终端运行的 coding agent 和 CLI，用于专业软件工程与复杂 coding work。", "bullets": [ "Grok Build 支持 plan mode：复杂任务先生成计划，用户可批准、评论单步或重写计划；批准后变更以 clean diff 方式展示。", "官方页面明确 AGENTS.md、plugins、hooks、skills 和 MCP servers 可直接工作，说明 xAI 选择兼容现有 agent repo 约定，而不是要求用户迁移到全新项目格式。", "页面还展示 parallel subagents、worktree integrations、headless mode `-p` 和 ACP support；这使它更像一个可脚本化 agent harness，而不仅是交互式 CLI。" ], "image_urls": [], "importance": "major", "editorial_category": "ai_industry" }, { "title": "OpenAI 发布 ChatGPT 在敏感对话中的上下文安全更新", "event_date": "2026-05-14", "url": "https://openai.com/index/chatgpt-recognize-context-in-sensitive-conversations/", "source": "OpenAI", "tier": "T0", "entities": [ "OpenAI", "ChatGPT", "Safety summaries", "GPT-5.5 Instant", "Sensitive conversations" ], "summary": "OpenAI 介绍 ChatGPT 在 sensitive conversations 中识别逐步显现风险的更新：系统会在少数高风险场景中使用短 factual safety summaries，帮助模型理解上下文并更谨慎地响应。", "bullets": [ "该工作聚焦 suicide、self-harm 和 harm-to-others 场景；OpenAI 称 safety summaries 不是通用 personalization 或 long-term memory，而是窄范围、限时、只在严重安全担忧相关时使用。", "OpenAI 公布的内部评估显示，长单轮对话中 suicide/self-harm 场景 intended safe response 提升 50%，harm-to-others 提升 16%；在 GPT-5.5 Instant 上，跨多对话 harm-to-others 提升 52%，suicide/self-harm 提升 39%。", "对构建面向高风险对话的产品团队来说，关键变化是 safety state 成为运行时上下文的一部分；这会影响日志保留、风险升级、用户透明度和普通对话误触发评估。" ], "image_urls": [], "importance": "major", "editorial_category": "ai_industry" } ], "projects": [ { "name": "Mux Robots usage update", "description": "Mux Robots 调整 workflow unit 计算并延长 free technical preview；本轮作为视频 AI 工作流商业化线索保留。", "url": "https://www.mux.com/docs/changelog/robots-pricing-update-may-2026", "domains": [], "importance": "general" } ], "builder_observations": [], "schema_version": 1, "canonical_url": "https://jasonxzwen.github.io/ai-daily-cn/reports/2026/05/2026-05-17.html", "html_path": "reports/2026/05/2026-05-17.html", "generated_at": "2026-05-16T18:42:58.656Z", "hero_highlights": [], "github_trending": [], "stories": [ { "story_id": "story-8-0-release-notes", "title": "ThreatConnect 8.0 发布 Agentic Threat Intelligence Platform", "importance": "major", "trend": "AI industry", "event_date": "2026-05-16", "primary_entity": "ThreatConnect Release Notes", "event_type": "launch", "object": "ThreatConnect 8.0 发布 Agentic Threat Intelligence Platform", "what_happened": "ThreatConnect 8.0 release notes 显示，该版本加入 Agentic AI beta，把威胁情报查询、STRIDE 威胁建模、情报报告生成和 Intelligence Requirement 创建做成四类可从 Ask AI 入口调用的代理能力。", "why_it_matters": "ThreatConnect Query Agent 会给出 source links 供继续核验；STRIDE Threat Modeling Agent 可从 Ask AI 或 Threat Actor Profile 详情页生成结构化报告。", "evidence_level": "primary", "sources": [ { "label": "ThreatConnect Release Notes", "url": "https://knowledge.threatconnect.com/docs/8-0-release-notes", "type": "source" } ] }, { "story_id": "story-aws-partner-central-agents-oppo", "title": "AWS Partner Central agents 支持用自然语言创建机会并提供 MCP 入口", "importance": "major", "trend": "AI industry", "event_date": "2026-05-15", "primary_entity": "AWS What's New", "event_type": "update", "object": "AWS Partner Central agents 支持用自然语言创建机会并提供 MCP 入口", "what_happened": "AWS 宣布 Partner Central agents 现在可通过自然语言对话加速 opportunity creation；该能力基于 Amazon Bedrock AgentCore，面向合作伙伴的 co-sell pipeline、deal hygiene 和 funding 机会识别。", "why_it_matters": "入口同时覆盖 AWS Console 中的 Amazon Q chat 和可编程 Model Context Protocol，因此销售团队可以从既有工具中触发机会创建，而不是只能进入 AWS Console 表单。", "evidence_level": "primary", "sources": [ { "label": "AWS What's New", "url": "https://aws.amazon.com/about-aws/whats-new/2026/05/aws-partner-central-agents-oppo/", "type": "source" } ] }, { "story_id": "story-changelog", "title": "Firecrawl v2.10 合并发布 parse、Lockdown、Question/Highlights 和新 SDK", "importance": "major", "trend": "AI industry", "event_date": "2026-05-15", "primary_entity": "Firecrawl Changelog", "event_type": "launch", "object": "Firecrawl v2.10 合并发布 parse、Lockdown、Question/Highlights 和新 SDK", "what_happened": "Firecrawl v2.10 changelog 把近几周的数据提取和安全功能合并到一个版本：`/parse`、Lockdown Mode、Question format、Highlights format，以及 Go、Ruby、PHP、.NET 官方 SDK。", "why_it_matters": "Lockdown Mode 允许 `/scrape` 只从 Firecrawl index 返回结果，不发起 outbound requests，并默认 zero data retention；该能力覆盖 API、CLI 和 MCP。", "evidence_level": "primary", "sources": [ { "label": "Firecrawl Changelog", "url": "https://www.firecrawl.dev/changelog", "type": "source" } ] }, { "story_id": "story-grok-build-cli", "title": "xAI 发布 Grok Build 终端 coding agent early beta", "importance": "major", "trend": "AI industry", "event_date": "2026-05-14", "primary_entity": "xAI News", "event_type": "launch", "object": "xAI 发布 Grok Build 终端 coding agent early beta", "what_happened": "xAI 宣布 Grok Build early beta，面向 SuperGrok Heavy subscribers 提供一个直接在终端运行的 coding agent 和 CLI，用于专业软件工程与复杂 coding work。", "why_it_matters": "官方页面明确 AGENTS.md、plugins、hooks、skills 和 MCP servers 可直接工作，说明 xAI 选择兼容现有 agent repo 约定，而不是要求用户迁移到全新项目格式。", "evidence_level": "primary", "sources": [ { "label": "xAI News", "url": "https://x.ai/news/grok-build-cli", "type": "source" } ] }, { "story_id": "story-chatgpt-recognize-context-in-sensitive-conversations", "title": "OpenAI 发布 ChatGPT 在敏感对话中的上下文安全更新", "importance": "major", "trend": "AI industry", "event_date": "2026-05-14", "primary_entity": "OpenAI", "event_type": "update", "object": "OpenAI 发布 ChatGPT 在敏感对话中的上下文安全更新", "what_happened": "OpenAI 介绍 ChatGPT 在 sensitive conversations 中识别逐步显现风险的更新：系统会在少数高风险场景中使用短 factual safety summaries，帮助模型理解上下文并更谨慎地响应。", "why_it_matters": "OpenAI 公布的内部评估显示，长单轮对话中 suicide/self-harm 场景 intended safe response 提升 50%，harm-to-others 提升 16%；在 GPT-5.5 Instant 上，跨多对话 harm-to-others 提升 52%，suicide/self-harm 提升 39%。", "evidence_level": "primary", "sources": [ { "label": "OpenAI", "url": "https://openai.com/index/chatgpt-recognize-context-in-sensitive-conversations/", "type": "source" } ] } ], "huggingface_trending": [], "model_releases": [], "hot_blogs": [], "chinese_media_dynamics": [], "daily_tracking": [], "official_org_updates": [], "evidence_assets": [], "quality_status": { "status": "ok", "public_note": "Core discovery checks completed without blocking degradation." } }