整理 | 郑丽媛出品 | CSDN（ID：CSDNnews）过去几年，AI 圈一直在疯狂讨论“大模型能力边界”。但很多人忽略了一件事：真正危险的，未必是模型本身，而是那些把模型连接到真实世界的基础设施。当 AI Agent 开始接管邮箱、数据库、企业 SaaS、代码仓库、云资源，甚至工业设备时，一个原本看起来“普通”的 Web ...

在过去的几年里，人工智能（AI）领域对于“大模型能力边界”的讨论如火如荼。然而，许多人却忽视了一个更为棘手的问题：真正的安全隐患，往往并非来自模型本身，而是那些将模型与现实世界连接的基础设施。当AI代理（AIAgent）开始掌控邮箱、数据库、企业SaaS平台、代码仓库、云资源，甚至工业设备时，一个看似普通的Web框架漏洞，可能瞬间演变为现实世界的安全灾难。 最近，安全研究人员揭露了一个令人不安的漏 ...

安全研究人员警告称，全球数百万个AI Agent和工具正受到一个严重漏洞的威胁，该漏洞可能允许黑客入侵运行它们的服务器，并窃取敏感数据和第三方账户凭证。该漏洞存在于Starlette中，这是一个开源框架，其开发者称每周的下载量为3.25亿次。数千个其他开源项目也存在漏洞，因为它们需要Starlette才能工作。该框架是异步服务器网关接口（Asynchronous Server Gateway In ...

Microsoft's AI Toolkit extension for VS Code now includes a Tool Catalog that can scaffold a Python or TypeScript MCP server with the core transport and registration plumbing already set up. In ...

Millions of AI agents and tools around the world have been imperiled by a critical vulnerability that can allow hackers to ...

NEW YORK, March 09, 2026 (GLOBE NEWSWIRE) -- Datadog, Inc., (NASDAQ: DDOG), the AI-powered observability and security platform for cloud applications, today announced that its MCP Server is generally ...

OX Security confirmed arbitrary command execution on six live platforms and estimates 200,000 MCP servers are exposed. Here's how to audit your deployments.

The first MCP-compatible private programmable onchain routing layer for autonomous agents launches on Solana. Non-custodial, compliance-checked, built for AI. The future of AI-agent commerce cannot ...

As more organizations configure MCP servers to support agent-to-agent communication, upfront strategy, nonfunctional requirements, and security non-negotiables will guide safer deployments. One of the ...

New capability lets lenders deploy custom AI agents directly inside loan workflows Blend Labs Inc. is opening its lending platform to third-party AI agents, launching a new “Autopilot MCP Server” that ...