TLDR IT 2026-10-13

Microsoft says 78% of all AI users bring their own AI tools to work. They're trying to resolve workflow bottlenecks you don't even know about with tools you can't even see.

This 12-question assessment from BrainStorm shows you the missed opportunities and compliance gaps that result from ad-hoc AI usage. Take it to see:

• What's already exposed
• Your risk scores across 3 areas: approved use, information protection, and sustained practice
• The manual bottlenecks that you could have automated but didn't because you lacked visibility

📑 You get a 30-day custom plan to address gaps.

Take the assessment

Energy company Ørsted automated firewall change management using Nautobot, reducing some requests from weeks to around 10 minutes. The system incorporates policy validation, approvals, and audit trails, demonstrating how network automation can improve speed without sacrificing compliance.

Microsoft rolled back a configuration change after administrators were unable to access Exchange Admin Center on Monday. Access has been restored, with Microsoft reviewing the change and its monitoring after another deployment-related disruption.

Attackers reportedly manipulated refrigeration controllers across hundreds of French supermarkets, triggering defrost cycles and causing food spoilage. The incident highlights how compromises of operational technology and third-party managed systems can create immediate physical-world disruptions.

While consumer AI agents and frontier coding models capture current headlines and capital, the most significant future revenue battle lies below the frontier in mainstream enterprise knowledge work. As open-weight and cheaper cloud or on-device models become capable of handling bounded tasks, a massive market will emerge for everyday business applications. Winning this "Switzerland" layer requires navigating complex trade-offs in data sovereignty, local versus cloud inference, and seamless software integration.

Technology expertise is only part of the job: IT leaders also need financial judgment, operational understanding, clear communication, and the ability to build effective teams. This roundup emphasizes connecting investments to measurable business results and making deliberate trade-offs instead of allowing every request to become another backlog item.

Moving toward fully AI-generated software without human code review is feasible if supported by extreme constraints and automated testing, similar to arguments made for self-driving cars. Existing legal frameworks and liability models, which often target human drivers or developers when things go wrong, are poorly equipped to handle automated systems where no one actually wrote the code.

Anthropic has launched the Critical Infrastructure Defense Program to help protect operational technology environments in sectors like water systems, power grids, and transportation. The initiative partners Anthropic with security firms and manufacturers including CrowdShield, Dragos, Nozomi Networks, and Rockwell Automation to deploy its Claude AI models and threat research for vulnerability assessment and patching. Additionally, the company introduced the OSS Scanner service to provide automated vulnerability reports to open-source software maintainers.

Raspberry Pi's lightweight desktop operating system for PCs and Intel Macs has returned with a Debian 13-based release. It includes a Wayland desktop and Raspberry Pi Connect remote access, but requires a fresh installation and supports only 64-bit x86 hardware, not Apple Silicon Macs.

WebMCP is a crucial primitive for enabling AI agents to interact with the web through co-browsing models. By allowing sites to provide direct JavaScript tools with explicit schemas, WebMCP offers significantly faster and cheaper agentic interactions compared to traditional DOM or screenshot parsing. Furthermore, imperative WebMCP tools are necessary to complement declarative HTML semantics and effectively handle multi-step user tasks.

A look at why TCP/IP has survived decades of extraordinary growth while so many competing technologies disappeared. The argument centers on open standards, interoperability, and practical incremental improvements rather than requiring everyone to replace everything at once.

An agent-security researcher explains why vulnerability tracking must extend beyond CVEs to include agent permissions, prompt injection, shared identities, and interactions between tools.

TinyJS uses native operating-system webviews to build small JavaScript desktop applications without bundling an entire browser engine into each app.