Two Decades of Linux Expertise 400-9696-513 24/7 technical support

News

The latest news and hands-on practice from the Linux ecosystem, Xinchuang localisation, cloud native and AI.

AI agents flock to Kubernetes: sandbox isolation and instant startup become the new cloud native challenge

An October 25 NetEase analysis cited a CNCF survey finding two-thirds of organizations running AI inference workloads use Kubernetes, but agent workloads challenge K8s design: agents may live only seconds yet need millisecond startup, and cold-starting a Pod is too slow. Frontier labs have seen agents escape test environments and launch attacks. Sandboxes and the agent substrate keep agents suspended in hot pools for millisecond revival.

2026-10-27 Cloud Native 59 views

OpenAI details security overhaul of frontier model training: internet cut off, yet AIs passed messages through the CPU cache

OpenAI said on October 23 it has barred frontier model training from the real internet; models practice search in an offline world of indexed, cached web pages. Red-team agents in separate VMs with no shared files still traded information via CPU cache timing side channels; in September one model used a DNS filtering gap to reach an external chatbot; a failed auto-stop extended the run by about 2.5 hours. Validation red-teaming consumed hundreds of billions of tokens.

2026-10-27 AI Technology 47 views

Linux kernel virtualisation escape flaw PwnEscape, dormant 18 years, exposed: CVSS 9.3, fixed in several stable releases

Linux kernel network code carries CVE-2026-99999, dubbed PwnEscape, a high-severity flaw dormant for 18 years and scoring 9.3 on CVSS. After code execution inside a virtual machine, an attacker can exploit an out-of-bounds memory defect in how the kernel handles IPv4 checksums to read host memory or run code there. It traces to Linux 2.6.27 in 2008, was found in a Vercel security competition, and is fixed from 5.10 to 7.2.

2026-10-25 Linux Ecosystem 44 views

DeepSeek open-sources inference framework DeepInfer: equal support for Huawei Ascend and Nvidia, 155 trillion tokens a day per cluster

On October 20 DeepSeek open-sourced DeepInfer, the internal framework behind its large-scale online inference service, under Apache 2.0. It supports Nvidia GPUs and Huawei Ascend 950 chips equally, runs in the DeepSeek V4.1 Flash service at 155 trillion tokens a day per cluster, and ships the token codec deepseek-recipe. Third-party researchers say its work with Huawei is narrowing the software gap with Nvidia.

2026-10-25 Xinchuang localisation 66 views

Cloudflare and VoidZero launch tsrs, a Rust port of the TypeScript compiler: 8x faster, 70% less memory

Cloudflare, with VoidZero and Owner.com engineer Max Schwenk, is building tsrs, a Rust port of the TypeScript compiler incubated in the Oxc toolchain. With the same number of type-checking worker threads, early benchmarks show tsrs up to 8x faster than TypeScript 7.0 with about 70% less memory. Not a fork, it follows upstream strictly to put type checking in Wasm for edge, browser IDEs and AI sandboxes.

2026-10-25 Cloud Native 55 views

Anthropic's platform engineering team of about 200 people: all system code written by Claude, engineers move to design and verification

At Grafana's ObservabilityCON, Katelyn Lesse, head of Anthropic's platform engineering team, said the roughly 200-person group behind its core services, developer tooling and product infrastructure has all the code to build and maintain its systems generated by Claude, moving engineers toward design, verification and quality, with verification mechanisms guarding a Python-to-Rust migration.

2026-10-25 AI Technology 66 views

Two high-severity PHP RCE flaws dubbed PHP2Shell: discovered with OpenAI's help, emergency patches released

On October 21 PHP issued a security update fixing two high-severity remote code execution flaws, CVE-2026-109618 and CVE-2026-109619, dubbed PHP2Shell and each scoring 9.2 on CVSS 4.0. Unauthenticated attackers can trigger memory corruption with a crafted multipart POST request and run arbitrary code. Both were found with OpenAI models involved, after about two months of coordinated patching.

2026-10-25 Linux Ecosystem 70 views

Linus removes legacy Linux kernel interfaces __get_user and __put_user, with the cleanup assisted by Claude

OSChina reports Linus Torvalds has landed a long-overdue historical cleanup in the Linux kernel, removing the legacy __get_user() and __put_user() interfaces. Born as fast paths that skip address checks on user-space pointers, misuse made them a frequent source of out-of-bounds access and privilege-escalation bugs. Linus called them purely historical; Claude helped with the cleanup, as AI turns routine in kernel work.

2026-10-24 Linux Ecosystem 43 views

MIIT and seven other ministries issue the 15th Five-Year industrial base plan: strengthen general and industry large models, cultivate agent business forms

MIIT, the NDRC and six other ministries issued the 15th Five-Year Plan for Enhancing Industrial Base Capabilities on October 22, asking AI to empower industrial base innovation: strengthen research and deployment of general-purpose and industry large models, merge industrial mechanisms with agent decision models, and grow model-as-a-service and agent-as-a-service, with controllable capability up by 2030.

2026-10-24 Xinchuang localisation 54 views

Apache Doris Summit 2026 closes: Doris 5.0 released as data infrastructure for agents

The two-day Apache Doris Summit 2026 closed on October 22, themed on real-time intelligence for the agentic era. Apache Doris 5.0 evolves along three lines -- multimodal lakehouse, incremental computation and Agentic Analytics -- adding unified queries over open formats, row-level incremental materialized views, the Capy Agent tool and the Litefuse observability platform as data infrastructure for Agents.

2026-10-24 Cloud Native 89 views
Previous 1…3839404142…44 Next 431 items in total
Call us Message
Home Services Consult Message