其他社区与社会服务专员

由于 Gemini 暂时不可用，该来源被挂到最接近的岗位上，以便先出现在岗位时间线中。

由于 Gemini 暂时不可用，该来源被挂到最接近的岗位上，以便先出现在岗位时间线中。

The Model Context Protocol (MCP), an open standard connecting AI agents to software, received its largest update, transitioning to a fully stateless architecture to enable massive enterprise production deployments. This update, under the Agentic AI Foundation, addresses operational burdens, hardens authentication, establishes a 12-month deprecation policy, and graduates interactive server-rendered interfaces and long-running asynchronous tasks into official protocol extensions. The changes are crucial for scaling AI agents, enabling organizations to run MCP servers with standard load balancers and cloud-native DevOps tools, and improving security and developer experience.

A VentureBeat survey of 107 enterprises reveals a significant AI agent security gap: 54% have experienced an incident or near-miss, yet most agents share credentials and isolation controls are rarely used. Enterprises primarily rely on provider-native security tools, express high satisfaction, but many plan to replace their stack within a year, indicating a disconnect between perceived comfort and actual exposure. Spending on agent security remains low, and most organizations believe the AI defense-offense arms race is at best even.

Enterprises face a significant "agent security gap" with over half experiencing AI agent security incidents due to inadequate identity, isolation, and enforcement controls. Most organizations rely on provider-native security tools, express high satisfaction, yet plan to replace tooling within a year, indicating a disconnect between perceived and actual security effectiveness and a lag in dedicated security budgets.

Enterprise AI adoption is encountering significant hurdles, with a VentureBeat survey revealing 86% of GPUs run at half capacity or less, and companies deploying AI agents ahead of critical controls for identity, evaluation, cost, context, and orchestration. This leads to security incidents and operational inefficiencies, prompting enterprises to retrofit their systems and plan vendor changes across these control layers.

Anthropic大幅更新了Claude Design，将其从一个基本的原型工具转变为企业平台。主要增强功能包括：引入设计系统以实现品牌合规性，与Claude Code的双向集成以简化设计到工程的交接流程，提高代币使用效率，以及扩展到各种创意和开发工具的导出选项。此举旨在将AI更深入地嵌入到企业的S设计和编码工作流程中。

由于 Gemini 暂时不可用，该来源被挂到最接近的岗位上，以便先出现在岗位时间线中。

The financial services sector is facing a new wave of sophisticated cyberattacks that bypass traditional password protection and MFA, primarily through social engineering (vishing IT support for MFA resets) and OAuth token theft (e.g., via Kali365 phishing-as-a-service). Reports from CrowdStrike, FBI, and Verizon confirm a structural shift where vulnerability exploitation now leads credential theft as an initial access vector. Security directors are urged to re-evaluate their defenses, focusing on out-of-band verification for MFA resets, restricting device code flows in Entra ID, monitoring OAuth refresh token usage, and auditing Graph API access. The industry needs to rebalance security budgets away from legacy, login-only MFA towards token monitoring, session validation, and identity verification to address these new threats that are making existing security measures less effective.

Raindrop AI launched Workshop, an open-source tool enabling developers to locally debug and evaluate AI agents by streaming all traces to a local dashboard, facilitating error identification and autonomous code fixing. The tool offers a 'self-healing eval loop' and integrates with various programming languages, SDKs, and frameworks.