TLDR IT 2026-09-22

Three days built for the teams running networks and security at scale:

· 100+ sessions across four tracks, including Networking & Security: Beyond SASE and Platform & Governance: Connectivity Cloud

· 200+ speakers, including Cloudflare co-founders Matthew Prince and Michelle Zatlyn, Stanford's Dr. Fei-Fei Li, Arm CEO Rene Haas, and Vue.js creator Evan You

· A full day of hands-on Cloudflare University training, plus the Global Partner Summit and Cold Start pitch competition on Monday

· Passes from $595, with team pricing at $495 for groups of five or more

→ Register now

Jev, TypeSafe AI's new model for making structured software decisions instead of generating text, became the fastest-adopted model in Vercel AI Gateway history. Nearly 13% of paid teams used it within 24 hours, highlighting growing demand for specialized models that handle agent routing, verification, and workflow decisions.

Nvidia has launched DSX Ready, a program that qualifies power and cooling systems for large AI factories. The initiative aims to help data center operators deal with increasingly tight power, cooling, water, and grid constraints as AI infrastructure scales.

Traditional incident response steps like revoking OAuth tokens fail when an implant can dynamically replace its entire credential set and OneDrive command-and-control identity via a single remote task. Investigators should treat application registrations as the durable object of compromise rather than relying solely on token revocation, and hunt on the identity plane using cloud telemetry.

Data ownership is fundamentally about accountability for business meaning, rules, and quality rather than technical possession or storage. While different teams across an organization may store, move, or consume data, business stakeholders remain responsible for defining its rules and metrics. Roles such as data owners, data stewards, and data custodians are distinct in managing enterprise data.

Keeper generates strong passwords, logs you in instantly, and notifies you in real-time if your credentials appear in a breach. It also enables secure family sharing for Wi-Fi, streaming, and account passwords. Everything is encrypted so only you can access it. Now personal and family plans are 50% off. Get your savings

Fastly launched AI Runtime Control and AI Firewall to manage model access, spending, routing, and agent interactions from its edge platform. The company says machine-generated traffic now accounts for more than half of traffic across its network.

VMware has quietly stopped selling its distributed firewalls for SmartNICs, walking back ambitions to bring hyperscale networking models to on-premise private clouds. The effort involved complex microcode integration with hardware from AMD and Nvidia, but struggled with Intel chips and ultimately failed to attract paying customers as conventional NICs improved. While standalone SmartNIC sales have ended, VMware's Distributed Services Engine remains supported within its Cloud Foundation suite.

Presidio opened its PATH AI lab to customers, allowing companies to test workloads across private cloud, public cloud, and neocloud infrastructure before committing to production deployments. The lab includes Dell servers running Nvidia B300 and RTX PRO 6000 GPUs.

Sumo Logic upgraded its Data Pipelines product so enterprises can filter, transform, and route telemetry before it is ingested. Its goal was to reduce observability storage costs while preparing cleaner data for AI-driven operations.

Atlassian's Data Brewery is a scalable synthetic data engine designed to generate high-volume, product-shaped, and relationally correct datasets. The system addresses the challenge of producing realistic customer data for products like Jira and Confluence in a maintainable and extensible way for engineers.

Mobile security firm Zimperium has discovered a new AI-powered Android malware named RatHat that tricks users into downloading fake applications and granting accessibility permissions.

KAYTUS upgraded MotusAI to let enterprises run and govern agentic AI workloads on-premises while claiming token operating cost reductions of 30–50%.

To accurately evaluate an AI tool, teams should test with competent non-experts, score against internal ground truth data, and strictly track whether the model or the human solved the task.