Microsoft has shipped sqlsimtools, an open-source, free toolkit for SQL Server load testing and workload simulation, as a public preview. It bundles the native command-line tool sqlsim, the Python port sqlpysim, the browser-based sqlsim Studio, and a Workload Builder agent for AI coding assistants. Use it to stress-test SQL Server, Azure SQL, and SQL databases in Microsoft Fabric, replay Extended Events captures, and compare runs before and after a change. Below are the components, requirements, installation steps, and known limits of the preview.

Source

Red Hat has shipped fixes for CVE-2026-98432, the Linux IPv4 TCP flaw that left RHEL and Ubuntu unpatched when researchers demonstrated a path to root. RHEL now rates the vulnerability Critical at 9.8, up from 5.9. Ubuntu’s fixes are still in an opt-in testing channel, with no date set for regular updates.

Source

Google’s Gemini 4 Carbon, a frontier model for coding and cybersecurity work, scored 59 on Artificial Analysis’ Intelligence Index V5—one point ahead of Claude Opus 5.5. After Google’s limited rollout to cyber defenders, the new results add a broader comparison of the model’s performance, reliability, and cost. Carbon leads four of nine tests, but it does not top every category.

Source

Cloudflare has launched Zero Wire, a public vulnerability-intelligence feed inside Cloudflare Radar that helps security teams assess critical flaws in internet-facing software. Its briefings combine plain-language explanations, affected deployments, recommended steps, and links to technical evidence and vendor guidance. A built-in prompt can help teams use an AI agent for a read-only exposure assessment.

Source

Microsoft Entra ID is rolling out a preview that lets users change their password without entering the current one, after signing in with an approved strong method. The option appears in My Sign-Ins and can avoid a self-service password reset or help-desk call. The feature is off by default, and rollout is expected to reach all tenants by the end of November 2026.

Source

A reported Microsoft Windows AI PC roadmap maps devices from entry-level 8GB laptops to workstations for running advanced models locally. Its headline-grabbing 512GB figure refers to memory per GPU in a planned workstation-class system—not ordinary system RAM. The leaked roadmap is unconfirmed, and its hardware plans may change.

Source

Network appliance vulnerabilities take more than a year to patch at the median, while exploited flaws reached CISA’s Known Exploited Vulnerabilities catalog in a median of just three days in 2026 through September. Microsoft’s analysis of about 383,000 CVEs shows the widest exposure gap is on critical systems that are slow to update.

Source

Dell and Microsoft are combining Azure Local, SQL Server and Dell’s private-cloud automation to help enterprises run and manage AI workloads across on-premises and cloud environments. The expanded integration targets a persistent obstacle to production AI: getting fragmented data and infrastructure to work together. Dell is also extending the partnership into cloud storage and locally developed AI agents.

Source

OpenVDDK is an open-source library that lets backup and migration tools read VMware virtual disks. Built to replace VMware’s libvixDiskLib.so.8.0.3, it matches the original library’s interface and is designed to work with tools that expect VDDK. Its creator reports tests on four vSphere versions and use by independent testers in several migration workflows.

Source

Fireperf is an open-source network testing tool for measuring how firewalls, routers, and other stateful systems handle connections—not just how much data they can transfer. It is available in packages for several Linux distributions, including Debian, Arch Linux, RHEL-compatible systems, and openSUSE. Its standout features include tracking connection setup failures and running multiple clients through one server port at the same time.

Source

AI coding agents are installing deprecated Model Context Protocol (MCP) servers that can expose developer machines to serious vulnerabilities. In 98% of reviewed sessions where an agent suggested or installed a deprecated server, it appeared to choose the package from its training data. The servers have no forthcoming patches, and their use can evade standard software inventory and security scans.

Source

Open-weight AI models could handle 90% of enterprise agent workflow steps, with best-case costs as low as one-twentieth of using a frontier model throughout, Gartner analysts say. In practice, extra retries and longer outputs may reduce the savings to five or six times. Gartner’s forecast for 2030 puts the emphasis on keeping routine operational work—and the data it processes—under enterprise control.

Source

Pando is a Git repository backend that lets OpenAI’s engineers and coding agents work in the same monorepo. Now powering all development there, it keeps existing Git workflows intact while handling concurrent pushes and high-volume fetches. OpenAI reports nearly fivefold lower median push latency and close to a petabyte of Git data served daily.

Source

OpenAI has increased rewards in its Security Bug Bounty program, with exceptional critical findings now eligible for up to $172,000 when bonuses are included. The company says average midpoint rewards for medium, high, and critical findings have more than quadrupled. New incentives reward detailed reports, repeat researchers, and findings on newly launched features.

Source

Anthropic’s revenue climbed 47% quarter over quarter to $17 billion, but that’s a sharp slowdown from its previous 144% jump—and its net loss ballooned to $23.3 billion.

Source

A Windows infostealer called VOLTA-AUTODOXER uses AI to sort stolen personal data and identify what may be easiest to monetize. It assembles information from accounts and devices into a victim profile, then sends it through OpenRouter for analysis. Cisco Talos has not confirmed that the malware has been used in the wild.

Source

Criminals are using stolen corporate AI access keys to run up bills exceeding $100,000 a day, with some victims facing nearly $1 million in charges before shutting off access. Google Mandiant says the attacks, known as LLMJacking, now include high-volume model use and services that resell access through stolen credentials.

Source

Google says Gemini 4 Argon has rolled out globally for complex tasks, while Gemini 4 Carbon—announced two weeks later—is now being tested with the U.S. government and selected cyber defenders. The new rollout broadens Argon beyond its initial subscriber availability, covered in Google’s first Gemini 4 Argon launch. Carbon is designed for long-running engineering work and cybersecurity defense, but remains in pre-release testing.

Source

NVIDIA says unauthorized activity on its Brev cloud development platform between October 4 and October 20 may have exposed SSH keys used to access customer virtual machines. The company replaced the potentially affected keys and updated customer instances, but has not confirmed whether attackers accessed workloads or data.

Source