Security dominated today's Software section, with ruby2shell flaw in Rails and Sinatra and phpAfterShell in PHP's session handler both drawing Cloudflare WAF coverage. Google detailed how GKE API timeouts blocked around 800 developers from deploying to production, tracing them to a Linux kernel TCP bug. Vercel raised its function request body limit, and Grafana 13.3 brought shared dashboard variables into public preview. Other developments included GitHub making security advisory permissions generally available and Digital.ai's Release 26.3 adding governance for AI agents in delivery pipelines.
Cloudflare says ruby2shell (CVE-2026-132825) lets unauthenticated attackers corrupt memory in default Rails and Sinatra applications. Its WAF now blocks the attack for every customer, including Free plan users. Cloudflare recommends upgrading Ruby now and using the WAF to cover the gap until the upgrade is in place.
Because the flaw needs no login and affects default configurations, Ruby upgrades should be treated as urgent.
Cloudflare describes phpAfterShell as a critical flaw in PHP's default session handler that can lead to code execution. Its WAF blocks the flaw for all customers, including Free plans. The same release also added WAF rules for other PHP flaws fixed alongside it.
A code execution flaw in a default PHP component is serious for any application using standard session handling.
Google detailed how intermittent Kubernetes API connectivity failures on Google Kubernetes Engine blocked around 800 developers from deploying to production. The failures came from three interacting bugs spanning GKE, Private Service Connect, and the Linux kernel. The write-up walks through how the team traced the timeouts across those layers.
It is a useful reminder that intermittent cluster API failures can originate below the platform layer, in the kernel networking stack.
Vercel Functions now accept inbound request bodies up to 20MiB, up from 4.5MB, in public beta. Projects opt in with an environment variable that applies to the whole project in that environment, so teams can test on preview before enabling it in production. The larger limit requires functions to use the 4GB performance memory setting and applies to Node.js, Python, and Bun runtimes plus Routing Middleware. Edge runtime functions keep the 4.5MB cap and return 413 on oversize bodies.
Larger request limits remove a common workaround for serverless apps handling files or big payloads, though they require the higher memory setting.
Grafana 13.3 introduces shared variables that work across dashboards and adds Notebooks for recording investigations. Both features are in public preview. Self-managed installations need feature toggles to enable them.
Shared variables and recorded investigations target the repetitive setup and handoff work that observability teams do during incidents.
A buffer overflow in OpenSSL's EVP_EncodeBlock() function is tracked as CVE-2026-121481 and carries a CVSS score of 9.8. An OpenSSL pull request addresses the flaw. The available details do not cover affected versions or whether exploitation has been observed.
OpenSSL is widely deployed across servers and devices, so a critical memory flaw in its core encoding routine calls for prompt patching.
Google is placing Agent Substrate and gVisor, two components of its AI agent infrastructure, under vendor-neutral open-source governance at the Cloud Native Computing Foundation. The move moves both projects out of sole Google control. The available details give no timeline or adoption figures.
Securing agents at scale is becoming shared infrastructure work, and a neutral home could make these components easier for other vendors to adopt.
AWS Emulator is a free, CC0-licensed local stand-in for more than 200 AWS services, generated from AWS's public API models. AWS says agents wrote and maintain all of its code and publishes fidelity scores for each release, aimed at the test loops of coding agents. The documentation says it does not evaluate IAM, and practitioners are divided on trusting an emulator that no one has reviewed.
Teams testing agent-written code against AWS can run checks locally, but the IAM gap and unreviewed code mean results need verification before anyone relies on them.
Meta developed Triton Low-Level Extensions, or TLX, to give developers low-level execution control on GPUs within the high-level Triton programming model. The extension is designed to exploit advanced capabilities of modern GPU architectures. The available details do not say whether TLX is generally available or how it is distributed.
Getting peak GPU performance often requires dropping to hardware-specific code, and a sanctioned extension in Triton could make that control more portable.
Uber redesigned its Go-based translation cache to reduce its initial memory footprint by more than 90 percent. The hybrid design keeps in-memory keys and frequently accessed values while moving the rest to disk-backed storage, which lowers garbage-collection overhead and memory use. Uber reports a cache hit rate above 99.9 percent and p99 disk-backed miss latency below 100 microseconds.
The approach shows how to shrink memory-heavy caches in Go services without giving up latency, a problem many large services face.
Five memory corruption bugs in nginx, collectively called nginx2root, can let an unauthenticated attacker run code on a server. The disclosure explains how each bug works. It also explains why origin servers behind Cloudflare's proxy are shielded from the most serious of them.
Another set of unauthenticated code-execution flaws in one of the most widely deployed web servers raises the urgency for operators to patch and to verify their exposure.
Discord CTO Stanislav Vishnevskiy announced a rewrite of the company's desktop and mobile clients in Rust, moving away from Electron and React Native. The project uses Zed's GPU-accelerated rendering framework and removes layers of abstraction so the apps draw directly on the GPU, use native mobile views, and share a Rust core across platforms. Early results point to a significant reduction in RAM usage and boot time for chat, voice, and streaming.
A large consumer app with a huge user base is abandoning Electron, a notable signal for cross-platform desktop and mobile development.
Google is adding computer use to its Gemini app for macOS. The assistant can now work across desktop applications. The announcement came three days after Apple tightened its rules.
It moves an AI assistant from chat into direct operation of apps on a user's Mac, raising new capability and platform-policy questions for developers.
Enterprise Live Migrations from Azure DevOps Services to GitHub Enterprise Cloud with data residency is now generally available. Teams can move their repositories while they keep working. The release targets enterprises that need data residency during the move.
Large enterprises can shift off Azure DevOps without stopping development, which lowers a major barrier to consolidating on GitHub.
CircleCI announced new features in its Smarter Testing suite. Smarter Testing runs only the tests affected by a code change to speed up feedback. The new Dynamic Parallelism adjusts the number of machines for each run, aimed at cutting run time.
CI pipelines are becoming the checkpoint for AI-generated code, so faster and more targeted testing directly affects team velocity.
Two nginx vulnerabilities, CVE-2026-123633 and CVE-2026-123007, allow unauthenticated remote code execution. The flaws are described as HTTP/2 related. The issue was widely discussed on the same day it surfaced, and operators running nginx should check for patches and affected configurations.
Unauthenticated remote code execution in nginx puts a very widely deployed web server and reverse proxy at risk.
OpenAI spent approximately $1.2 million in API costs to generate working exploits for 948 Linux kernel vulnerabilities. Sam Altman says this is just the beginning. The effort shows how cheaply AI can now turn kernel bugs into working attacks, which raises the pressure on maintainers and distributions to patch faster.
It is a concrete cost benchmark for AI-driven vulnerability discovery against the software most of the internet runs on.
Java4Shell, tracked as CVE-2026-114938, is a memory-corruption flaw in OpenJDK's Base64 decoder that can lead to remote code execution. Cloudflare has released six WAF managed rules for paid plans, which log matching requests by default. Cloudflare advises patching Java runtimes now.
A remote code execution flaw in the OpenJDK runtime affects a large share of server-side Java deployments, so patching is the priority.
Vercel has released UI SDK in beta, an Apache 2.0 TypeScript library with more than 50 UI behaviors in @ui-sdk/core. It has adapters for React, Vue, Svelte, Solid and Web Components. Agents can compose accessible components and operate them through a documented agent bridge.
It gives agents a standard, accessible component layer across major frameworks, which is a practical foundation for agent-built front ends.
Shopify built Remote Dev, a remote mobile development environment. It lets AI agents independently build, test and maintain iOS and Android applications in parallel. The design keeps the resource load off developers' own machines.
It shows a production approach to running many AI development agents on mobile codebases at once, without local resource limits.
Microsoft has released .NET 11, a Standard Term Support release with two years of support. It ships C# 15 with union types and closed hierarchies, a Native AOT CLI host, and a preview of Runtime Async. ASP.NET Core gains Zstandard compression, asynchronous validation and OpenAPI 3.2, and .NET MAUI now uses CoreCLR by default. EF Core 11 adds experimental vector search.
It brings language, runtime and web framework changes to a widely used development platform, with a long support window for teams planning upgrades.
Vercel has announced eve 1.0, an open-source framework for building production-ready agents, now generally available after a public preview in June. The company says it shipped more than 230 releases and merged thousands of pull requests while preparing 1.0. The release adds faster performance, expanded integrations and improved observability, plus security controls that let developers decide which actions need approval and what data gets collected. Vercel also says it measured how well coding agents can build with eve using an evaluation suite.
A stable open agent framework with approval and data controls gives teams a production-oriented base for building agents.
GitHub released a batch of Copilot updates timed with Microsoft Ignite. The Copilot app now supports Azure DevOps repositories, GitHub Enterprise Server and the browser, Copilot code review adds adaptive effort and secret detection, and Copilot for JetBrains gains remote access and on-device voice dictation. Copilot CLI adds Microsoft Entra sign-in in public preview, Copilot agent activity can be streamed into Microsoft Agent 365, and GitHub Copilot Evals and on-demand secret scanning are available. MAI Code 2 and MAI Code 2 Flash are now in Copilot, and Microsoft claims near-Opus performance at 5% of the cost.
Copilot now reaches more development environments, adds governance and audit hooks for enterprises, and brings Microsoft's own coding models into the platform.
OpenAI researchers privately reported critical security vulnerabilities in PHP on August 11, 2026. The flaws affect PHP, which underpins millions of websites built on platforms such as WordPress and Drupal. Attackers targeted operational websites and critical systems using the vulnerabilities before the fixes were publicly released.
Widely deployed open-source infrastructure is being attacked during the window before patches are public, so site operators need to patch fast and treat AI-driven vulnerability discovery as an active threat.
In the Codename MDASH pilot, GitHub and Microsoft turned thousands of raw AI findings into 99 checked, private reports with suggested fixes for six widely used open source projects. Maintainers have already fixed more than half of them. The pilot's premise is that maintainers need fixes rather than more AI-generated bug reports.
It offers a model for AI-driven vulnerability discovery that reduces maintainer workload instead of adding to it across foundational open-source software.
AWS has launched Nitro Sandboxes, which run each opted-in EKS pod, ECS task or EC2 workload inside its own Firecracker microVM with a separate kernel. It is enabled by a single runtimeClassName field and carries no extra charge. A confidential mode adds attestation-gated key release. The documentation lists trade-offs: no GPU passthrough, at most five sandboxes per vCPU, and half the compute on SMT instances.
It gives Kubernetes and container users a stronger isolation boundary for multi-tenant workloads with minimal configuration change, though the GPU and density limits matter for planning.
A recently disclosed OpenSSH vulnerability, named *sshole and tracked as CVE-2026-123158, lets an unauthenticated attacker log into accounts that should have no usable password. It affects Linux systems running OpenSSH with the musl C library that do not use PAM. Security researchers now report that attackers are exploiting it in the wild. Alpine Linux users are being urged to patch.
Exploitation of an unauthenticated remote login flaw in OpenSSH is an urgent patching priority for anyone running affected Linux builds.
DuckDB's developers released what they describe as the largest version in the project's history, bundling several ideas for improving how the database stores data. The team has said the release should not yet be run in production. The release is aimed at changing the storage layer rather than adding small features.
A major storage change in a widely used embedded analytics database signals where its file format and performance are heading, even though it is not yet ready for production workloads.
Uno Platform 7.0 unifies rendering on SkiaSharp 4 with an opt-in WebGPU backend and adds Apple TV as a target. It supports .NET 11 from day zero and runs third-party .NET MAUI controls and Microsoft.UI.Reactor on all targets. The release removes native rendering and Mac Catalyst support.
Developers using Uno for cross-platform .NET apps must plan for removed Mac Catalyst and native rendering paths, while gaining a single rendering stack and new targets.
Security Journey announced on November 12 that it has launched integrations with Claude and ChatGPT built on the Model Context Protocol. Program owners can check on learners, assign training and pull reports in plain language from inside those AI assistants. This removes the need to log in to the training platform separately.
It is another example of a business software vendor exposing its core workflows as MCP tools inside the assistants that teams already use.
Security findings describe Lovable-built applications exposing passport details and more than 200,000 private messages. The item is a detailed research write-up that has drawn wide attention, with a discussion thread of several hundred comments. The excerpt does not give the number of affected apps or whether the issues have been fixed.
It is a concrete example of the data exposure risks that come with apps generated quickly by AI-assisted builder platforms.
Cloudflare has made Application Services for Private Origins generally available. The feature lets public traffic reach applications hosted on private networks without exposing their IP addresses or opening inbound firewall ports. Private-to-private connectivity is not yet supported.
It gives teams a way to publish private applications through the edge without opening inbound firewall paths, which is a common security and operations constraint.
Anthropic has described how the Claude Code team reviews code as AI agents increasingly contribute to its codebase. The approach uses several layers of automated verification, AI-powered reviews and risk-based human approval. The stated aim is to reduce manual review time without increasing regressions or production incidents.
It is a rare, detailed look at how a major AI coding tool vendor is governing agent-written code in production.
The Python Package Index has made its redesigned project pages live, with updated UI and clearer security signals on each package. The discussion accompanying the announcement drew over a hundred comments. The excerpt does not list the specific signals shown.
Package pages are where developers judge whether a dependency is safe to install, so clearer security information there affects supply-chain decisions.
DuckDB has published a response to older database protocols, which send results row by row and repeat details such as column names. Those protocols were designed before big data workloads. The excerpt does not give benchmark figures or details of the implementation.
Protocol overhead on large result sets can slow data pipelines, so a more efficient wire format matters to anyone moving large query results.